The Economics of Medical Procedure Innovation

THE ECONOMICS OF MEDICAL PROCEDURE INNOVATION

David Dranove Kellog School of Management Northwestern University 2211 Campus Drive, Evanston, IL 60208 Email: [email protected]

Craig Garthwaite Kellog School of Management Northwestern University 2211 Campus Drive, Evanston, IL 60208 Email: [email protected]

Christopher Heard Department of Economics Northwestern University 2211 Campus Drive, Evanston, IL 60208 Email: [email protected]

Bingxiao Wu* Department of Economics Rutgers University 75 Hamilton St, New Brunswick, NJ 08901 Email: [email protected]

* Corresponding Author. Acknowledgement: The authors declare no financial or personal relationship that could cause a conflict of interest regarding this article.

THE ECONOMICS OF MEDICAL PROCEDURE INNOVATION

Abstract

This paper explores the economic incentives of medical procedure innovation. Using a novel, proprietary dataset on billing code applications of emerging medical procedures, we highlight two mechanisms that could hinder innovation. First, the administrative hurdle of securing permanent, reimbursable billing codes substantially delays innovation diffusion. We find that Medicare utilization of innovative procedures increases nine-fold after the billing codes are promoted to permanent (reimbursable) from provisional (non- reimbursable). However, only 29 percent of the provisional codes are promoted within the five-year probation period. Second, medical procedures lack intellectual property rights, especially those involving no patented devices. When appropriability is limited, specialty medical societies lead the applications of billing codes. Our work indicates that the ad hoc process that oversees procedure innovations creates uncertainty over both the development process and the allocation and enforceability of property rights; this stands in stark contrast to the more deliberate regulatory oversight for pharmaceutical innovations.

Key Words: Medical Procedure Innovation, CPT Codes, Property Rights JEL: I11, O33, I18

1. Introduction

Improvements in medical technology have been a primary driver of increased life expectancy and

medical spending (Cutler, 2004; Newhouse, 1992). These improvements have taken many forms. Society now

enjoys access to pharmaceuticals that treat a wide range of both common and rare conditions. For example,

medications exist to help lower blood pressure and cholesterol levels, cures have been developed for hepatitis

C, HIV has been transformed from a horrific plague into a largely manageable condition, and a variety of

gene therapy products are promising cures for rare illnesses that previously served as death sentences.

Technological progress has also been made in medical procedures. Surgical improvements now allow medical

providers to address a wide range of medical conditions including, but certainly not limited to, relatively non-

invasive surgeries for heart attacks, improvements in the diagnosis and treatment of strokes, surgical solutions

for various types of cancer, and a variety of effective mental health treatments.

A rich economics literature has evolved examining how firms decide to invest in developing medical innovations. This literature, however, has primarily focused on both investments in pharmaceutical innovations and how the resulting drugs diffuse into clinical practice.1 This focus is likely driven by the fact that the economic model and the regulatory process governing drug development are more clearly understood and data about each stage of the production process are more widely available.

In contrast, economists have devoted much less attention to the development of new medical procedures.2 As a result, little is known about the process by which such novel procedures are developed let alone whether this process is optimal. In this paper, we attempt to fill this gap in the existing literature by examining the underlying economics, rules, and regulations governing the innovative process for medical procedures. Whereas every step of pharmaceutical development is planned and regulated, medical process improvement comes about in a far more ad hoc manner. This includes both the regulatory structures and the

ways in which products eventually diffuse throughout the medical community.

1 See, e.g., Acemolgu and Linn, (2004), Finkelstein, (2004), Blume-Kohut and Sood, (2013), Dubois et al. (2015), Agha and Molitor, (2018), Dranove et al. (2020), and Garthwaite et al. (2020). 2 There is a small literature on medical devices, as we describe below. However, many new value creating procedures are not directly connected to new devices.

The broad economic decision is the same for both products and procedures. In both categories, potential innovators must make large, fixed, and sunk investments in research and development. Such investments can take the form of explicit financial investments as well as the opportunity cost of medical research teams. Despite this fundamental economic similarity, and the potentially large welfare gains afforded by both types of technologies, the regulatory and legal frameworks governing the development of these two types of innovation are vastly different. Such differences can influence both the amount and scope of resulting innovation.

The development of pharmaceuticals is heavily regulated by both the Food and Drug Administration

(FDA) and the patent system. This accomplishes two goals. First, the FDA approval process assures a measure of safety and efficacy. Second, firms are granted clear and well-understood protections over the intellectual property resulting from their successful investments. In addition, a variety of other regulations and competitive forces cause payers to cover nearly all FDA-approved pharmaceutical products. The combination of these factors means that innovative firms have some certainty they will maintain pricing power resulting from their market exclusivity for the novel pharmaceutical products. Policymakers and economists alike believe that without the market power created by this market exclusivity, firms would underinvest in the development of new technologies (Nordhaus, 1969). A robust empirical literature supports this belief by demonstrating that increased market opportunities drive investments in research and development.3

Fundamentally, these policies represent a tradeoff of accepting decreased welfare from reduced access to products today in order to provide the incentives for firms to make the investments necessary to develop goods and services in the future.

There are at least two primary differences between the development process for drugs and

procedures that should directly affect appropriability and therefore the firm’s optimal investment decision.

The first is a question of property rights. The development of novel pharmaceuticals involves well-defined

property rights that are enshrined in the patent system. As a result, firms making large and sunk investments

3 For example, see Ward and Dranove (1995), Acemoglu and Linn (2004), Finkelstein (2004), Blume-Kohout and Sood (2013), and Dranove et al. (2020).

4 in developing new drugs have a reasonable assurance they will capture a meaningful portion of the value created by successful products. The same cannot be said for the firms and individuals who develop new procedures, for several reasons. First, procedures are difficult to successfully patent.4 Second, many new procedures involve devices that no longer have exclusive patent protection, either because the patent has expired or competing devices have reached the market. Even when patent-protected devices are involved, the financial returns of a new procedure are diffused across both the device maker and a large number of providers of the procedure. For example, while stents can cost anywhere from a few hundred to a few thousand dollars, the median facility fee for stent placement is $21,000.5 Even after provider discounts, the facility payment dwarfs the payment to the stent manufacturer. Much of this payment represents incremental profits above variable costs. Surgeons’ fees are also in the thousands of dollars. This stands in contrast with new drugs, whose prices, and the resulting profits enjoyed by their patent holders, often vastly exceed the fees paid to providers who prescribe them.

A second important difference is that novel pharmaceutical products approved by the FDA receive almost immediate acceptance and reimbursement by payers. In contrast, the developers of new procedures have far less certainty about whether, when, and how much they will be reimbursed. As we explain below, reimbursement critically depends on whether and when the American Medical Association (AMA) grants a

Current Procedural Terminology (CPT) billing code, an administrative process that has heretofore been relatively obscure to most researchers.

We begin by focusing on the former of these two differences using a novel dataset from the AMA

that contains the universe of CPT applications. These novel data allow us to identify the entirety of the

development process for new procedures, including the first benchmark research, the FDA approval of

related devices, the AMA approval of “temporary” CPT III codes (for which insurers generally do not

reimburse), and the subsequent promotion to “permanent” CPT I codes (for which insurers nearly always do

4 See Section 5.3 for details. 5 See https://www.epainassist.com/test-and-procedures/cost-of-stent-placement-procedure. Accessed 8/12/2020.

5 reimburse). Given that most new medical procedures involve medical devices, this descriptive information complements and extends the existing evidence in this area (e.g., Stern 2017).6

Among our key findings, we demonstrate that the timeline for discovering, developing, and commercializing novel procedures is far longer than previously reported. This difference primarily results from the fact that previous efforts focused on only specific stages of the development process and/or could not observe the administrative process determining reimbursement.7 We document an average lag of more

than six years between initial research publications about procedures involving new devices (i.e., those that

have not previously been used in another medical procedure and have been approved less than two years

before CPT III application) and filing of the FDA application. We then document an additional one year

before FDA approval.

If this were a pharmaceutical process, then this seven-year-long development period would largely

describe the time period before innovative firms could expect to earn revenue from a successful innovation.

It would also support the common belief that procedures and devices could be developed more quickly than pharmaceuticals, with the latter taking nearly eight years on average from the initial clinical testing to

marketing approval (DiMasi et al. 2016), even longer if we start the clock with the granting of patents.

However, procedures face additional hurdles before they are widely reimbursed and diffused throughout the

medical community. After FDA approval of the device, we estimate there are an additional nine months until

the awarding of CPT III status of the procedure codes, and even these provisional codes are rarely reimbursed. Furthermore, only a small fraction of procedures subsequently advance from CPT III to fully reimbursed CPT I status—among all CPT III procedures approved between 2008 and 2014, only 29 percent are promoted after the five-year temporary period.8

6 We offer one important caveat. While there have been significant welfare gains from incremental improvements to existing procedures – consider for example the tremendous improvements in outcomes for CABG surgery patients since the first procedures were performed 60 years ago (Goetz et al. 1961)– our focus is on new procedures. We will return to this distinction in our conclusions. 7 For example, Stern (2017) documents that the FDA requires less than two years after submission, on average, to review and approve new Class III (“High Risk”) medical devices, while Makower et al. (2010) report that the average time between first communication and FDA approval is 54 months. A high risk device “supports or sustains human life, is of substantial importance in preventing impairment of human health, or presents a potential, unreasonable risk of illness or injury.” Source: https://www.fda.gov/medical- devices/premarket-submissions/premarket-approval-pma. Accessed on 8/23/2020. 8 For procedures involving new devices, only 17 percent are promoted within five years, and the average lag for promoted codes is 34 months.

Combining all stages of development together, we find a total average time from initial research to

CPT I status for procedures involving new devices is about ten years, as long or longer than the timeline for new drugs.9 In addition, this process contains additional forms of uncertainty about whether even successful innovations will be reimbursed by payers and/or implemented by providers.10

We also document that many new procedures involve devices that had previously been approved for other procedures. While these devices no longer require FDA approval, the associated new procedures may require new CPT codes, depending on whether they are substantially different and/or require substantially different effort by providers. We find that the CPT approval process for these “old device procedures” is

even longer and less certain than for new devices. The same is true for new procedures that do not involve

devices. While providers may be able to bill for the new procedure using an existing CPT code, this is far

from certain. Even if insurers do reimburse under an old code, the payment may not fully compensate for the

effort required to perform the new procedure.

A primary contribution of our results is demonstrating that the importance of the under-discussed

administrative hurdle of securing a billing code to the development process for new medical procedures – a

process, we note, that is largely absent from the calculations of firms investing in the development of new

drugs.11 We demonstrate that these codes are not simply the administrative formality that their absence from

the literature suggested. We estimate that the AMA’s decision to promote a code from CPT III to CPT I

causes a statistically and economically significant increase in the use of these procedures by Medicare patients.

This stands in stark contrast to new drugs, where no third party “seal of approval” beyond the FDA is

required for firms to begin earning revenue. Therefore, the considerable lag between FDA approval and

promotion from CPT III to CPT I codes represents a meaningful economic cost for innovators. To the

9 The innovation time is much longer for procedures involving no device or old devices. See Section 3 for details. 10 As we explain below, some of the delay may reflect the fact that providers may be able to bill for these procedures using existing CPT I codes, so there is less urgency in seeking new codes. In fact, given the limited reimbursement for CPT III codes and the lack of certainty about moving to CPT I, in expectation it may be more profitable to remain being reimbursed under the original code. 11 Drug makers are not guaranteed reimbursement after FDA approval. Normally, they must negotiate with Pharmacy Benefits Managers (PBMs) for inclusion in formularies. PBMs are increasingly moving to closed formularies that do restrict coverage. See, e.g., Agha et al. (2020).

7 extent that the procedure requires a patented medical device, this delay likely affects particularly valuable periods of market exclusivity.

We next use our novel CPT application data to shed light on the potential role of property rights in the development process of medical procedures. For pharmaceutical products, firms that bring a product to market receive patent protection and are the firms that apply to the FDA for regulatory approval. For medical procedures, however, the firms or individuals that develop the procedure only receive defensible property rights to the extent that the procedure involves a patentable medical device. Even in that case, the property rights only pertain to the device which represents a fraction of the spending on many new procedures. This limits the appropriability of investments in new medical procedures.

At the margin, this limited appropriability could decrease the level of investment in procedures compared to pharmaceuticals that generate similar societal value. There could conceivably be rational reasons to prefer one category of innovation over another. That said, allowing less appropriability for procedures compared to pharmaceuticals does not appear to be a well thought out policy decision. Nor, does it necessarily reflect that these two product categories generate inherently different levels of economic value. In fact, many procedures have meaningfully affected the provision of medical services. Consider the transcatheter aortic valve replacement (TAVR), which was a revolutionary and lifesaving treatment for many patients with diseased aortic valves. Similarly, Applied Behavior Analysis (ABA), which is a set of related assessments and treatments for producing “clinically significant and lasting improvements in the functioning of people with autism.”12 Both procedures are extremely popular, provide substantial benefits to patients, and generate economically meaningful medical spending. For example, Medicare beneficiaries received over

50,000 TAVR procedures in 2016, and the adoption of TAVR has been described as a “tsunami” (Leon et al.

2018). ABA is sufficiently widespread to have its own board that has certified over 33,000 behavior analysts.13

Even beyond these well-identified cases, a casual perusal of the practice of medicine in hospitals over the last

30 years would demonstrate that procedural innovation has had a meaningful effect on this sector.

12 We purposely avoid calling these “process innovations,” a term often used to describe an innovation that allows one to produce an existing product at a lower cost or higher quality. 13 Source: Behavior Analyst Certification Board (BACB) 2019. BACB Certification Data. BACB. https://www.bacb.com/bacb- certificant-data/ Retrieved on 1/12/2020.

We examine the potential role of appropriability in the development of new procedures by examining the sponsors of applications for new CPT billing codes. Broadly speaking, applicants in our data are either firms or professional medical societies. We document meaningful differences in the types of procedures supported by these two types of applicants. For industry participants, 92 percent of CPT III applications are for procedures that involve an exclusive patented device.14 In contrast, only 36 percent of medical society

applications involve such a device. Furthermore, those remaining procedures supported by industry

applicants that involved generic devices were promoted by consulting firms that are in the business of

assisting physician groups with reimbursement. While not dispositive, the clear pattern in the data suggests

that private firms are less likely to support devices without a clear path to appropriability. This could mean

that process improvements unrelated to devices are underprovided to the market.

2. Background

2.1. The Procedure Innovation Process

At a broad level, the scientific process leading to medical procedure innovation resembles the more often discussed process for discovering and developing new drugs.15 For pharmaceuticals, target molecules

are identified and refined over time through a combination of theoretical and laboratory research with limited

clinical applications. Innovators often publish their findings throughout this process, both to help establish

precedence and to provide information to the medical community. At some point, usually early in the

discovery process, innovators almost always seek patent protection which provides clear intellectual property

protection against other firms commercializing the target molecule.

In order to bring a drug to market, firms must then apply for permission to participate in the multi-

stage drug development process, overseen in the United States by the FDA. Following FDA approval,

insurers nearly always pay for the drug, although the exact amount is sometimes subject to negotiation.

14 We define such a procedure as one that a) involves a medical device, b) the medical device is made by only one firm (i.e., no competing firms) based on the referenced studies in the CPT III application, and c) the medical device company has an unexpired patent claiming the device at the time of CPT III application. 15 Several studies document drug development times, success rates, costs, the FDA approval process and the timelines involved. See, e.g., Dranove & Meltzer (1994), DiMasi et al. (2003), Keyhani et al. (2006), Hay et al. (2014), DiMasi et al. (2016), and Wong et al. (2019).

Finally, providers and patients decide whether to use the new drug based on the available evidence and the

relevant reimbursement decisions.

This patent and approval process is highly standardized. The state of patent law for new drugs is

mature and the FDA provides clear guidelines about acceptable evidence and criteria for approval. Firms still

face scientific risk. After all, clinical trial outcomes are unpredictable and failure to successfully progress through all phases is relatively common (Wong et al. 2019). However, the FDA has a long history of systematic decision making and drug makers have established relationships with the FDA. Therefore, conditional on generating sufficient scientific evidence of safety and efficacy, firms have general predictability concerning the lengthy regulatory review process. Reimbursement of new drugs is common, typically follows shortly after FDA approval, and is largely based on market prices (Newhouse 2004).

The regulatory process for procedures exhibits far more variation. The closest match is for procedures involving new medical devices. In these cases, the FDA plays an important regulatory role, which has been described by Stern (2017) so we will only present the essentials necessary for understanding the economic issues discussed in this paper. The FDA review of devices varies according to the type of device.16

Class I and Class II devices are both considered low risk, although Class II devices require special standards

of care to assure patient safety. Approval of both Class I and Class II devices requires relatively little scrutiny.

Makower et al. (2010) report that the average time between a low-risk device maker’s first communication

with the FDA and approval was just 31 months.

Firms that wish to market high-risk devices (i.e., Class III devices) must file for premarket approval

(PMA). The PMA must provide evidence of safety and effectiveness, normally derived from clinical trials.

Stern (2017) provides a systematic review of the PMA review process and documents that conditional on the

submission of FDA applications, the approval rate is very high. Makower et al. (2010) report that the average

time between first communication and FDA approval is 54 months. Stern (2017) finds that the average FDA

review time is 18 months, suggesting that the clinical trials require three to four years.

16 See Goldman & Lakdawalla (2011).

Some innovations represent applications of existing devices to new indications, exactly analogous to

drugs. Unlike drugs, however, FDA approval for each new indication of an existing device is more rapid,

often a matter of months.

Finally, there are some new procedures that do not involve any medical devices. For these

procedures, the FDA plays no regulatory role in the development, testing, or marketing. In principle, there is no direct regulation of new procedures of this type. Innovative providers can experiment in their development and other providers can simply adopt them at their discretion (Darrow 2017). Adoption and diffusion are highly dependent on education within the medical community, such as through presentations at meetings of professional societies (McKinlay 1981). Beyond education, adoption also depends on reimbursement – a process we detail below.

2.2. Administrative Coding and Reimbursement of Medical Procedures

Medical procedures are complex and heterogeneous products, so payers have adopted standardized

coding systems for different procedures as the administrative basis of reimbursement. Each service is reported to payers using the most appropriate code and reimbursed accordingly. Unlike drug treatments that are standardized by regulatory bodies (one dose must be chemically identical to any other), medical procedure coding inevitably groups slightly different services together under a single code. Furthermore, insurers often base payments for medical procedures on estimated costs, unlike drugs where prices are largely market-

based.17 Specifically, Medicare uses a “prospective payment system” for both inpatient and outpatient

procedures, where payments are based on a standardized measure of costs per procedure. A large percentage

of private payers follow Medicare, paying providers a multiple of Medicare’s rates.

The willingness of providers to adopt new procedures relies on a procedure being sufficiently

profitable or the provider being willing to undertake the procedure at a loss in order to satisfy some other

part of their personal objective function. This means that, within the medical coding framework, one of three

situations must apply. In some cases, a new procedure is sufficiently similar to existing procedures that

17 Public payers use explicit cost-based payment structures, such as Diagnosis Related Groups (DRGs) and Relative Value Units (RVUs). Private payers often a multiple of DRG case weights and RVUs. Thus, even in a market setting, payments are tied to costs.

providers may bill under an existing code. This will persist as long as the new procedure is profitable at the

rates payers are willing to pay for that code. In the second situation, the new procedure has a new code, with

a corresponding new reimbursement rate. This is likely to arise when the new procedure is sufficiently

different from existing procedures to make it impossible to bill under an existing code and/or the new procedure is not profitable at rates paid for the old procedure. Finally, providers can bypass the standard medical coding system and request ad hoc reimbursement from payers. This may occur for one of two reasons.

First, there may be no similar procedure with a CPT I code. Second, there may be a similar procedure, but providers believe they can receive higher reimbursement if they request ad hoc payment. Either way, billing outside of the standard system can be administratively onerous and risky for providers because there is generally no guarantee of reimbursement before treatment and payers can be reluctant to accept ad hoc claims.

Adding further complexity is the fact that different coding systems are used in different situations.

Two of the most important are the Current Procedural Terminology (CPT) code set maintained by the AMA,

and the Medicare Severity-Diagnosis Related Group (MS-DRG) codes maintained by the Centers for

Medicare and Medicaid Services (CMS).18 The same treatment might be reported using multiple code sets because there are multiple providers to be paid. The situation under Medicare is informative. In outpatient settings, Medicare (and most private payers) uses CPT codes as the basis for fee schedules. In inpatient

settings, Medicare uses both CPT and DRG codes.19 Private payers use a wider variety of codes, including but not limited to CPT and DRG codes.

There are two important implications of this discussion of medical coding for the adoption and diffusion of innovative medical procedures. The first is that some utilization of innovative procedures may

not be clearly reflected in administrative data. Unlike new drugs that must be reimbursed under their (new)

names and are easily tracked, procedure innovations that are claimed and reimbursed under existing

procedure codes (before the new codes get approved) will be difficult to detect in claims data. We explain below how we address this issue by taking advantage of the availability of the provisional (Category III) CPT

18 The CPT and MS-DRG codes are the bases for other code sets. For example, the CPT code set is the core of the Healthcare Common Procedure Coding System (HCPCS) used by Medicare. 19 CPT codes are used for physician reimbursement. DRG codes are used for hospital reimbursement.

codes in our data. The second is that for procedures that do require new codes, the new code assignment may

be an important part of the innovation process that has not previously been studied and has no analogy to

drugs. In Section 2.4, we provide more details about this process.

2.3. The CPT Code Approval Process

The AMA maintains and holds the copyright to the CPT code set described above. The AMA holds several meetings each year to consider changes to the code set, including adding new codes, removing old codes, reallocating or consolidating existing codes, and refining code descriptions. Understanding the process by which new codes are assigned requires an understanding of the structure of CPT codes.

There are three types of CPT codes. Category I (CPT I) codes form the bulk of the 10,000 CPT codes representing virtually all procedures that medical science has to offer. A CPT I code contains five digits organized by specialty or type of service. For example, codes 00100-01999 pertain to anesthesia while 99201-

99499 are for evaluation and management services. CPT I codes are effectively permanent.20 Category II

(CPT II) codes are optional ‘add-on’ codes that providers can report alongside other procedure codes and are used for execution and performance measurement. They are not relevant to our study. Starting in 2001, the

AMA introduced Category III (CPT III) codes, which contain four digits followed by a “T”, such as 0099T

(implant corneal ring). CPT III codes are temporarily assigned to procedures that do not meet the criteria for

CPT I codes. They are designed to track the use of emerging procedures while evidence accumulates about

whether a CPT I code should be assigned.

Finally, a class of CPT codes is reserved for procedures that do not appear elsewhere in the code set.

These codes typically end with “999” or “99”. For example, code 33999 covers unlisted cardiac surgery

procedures A provider using these codes typically submits a request for reimbursement, describing the

procedure, the medical justification, and the requested payment.21 Each payer determines whether and how

much to reimburse the provider for each code. For a common new procedure that is awaiting AMA approval,

such as TAVR, providers may systematize the process of requesting reimbursement, and payers may routinely

20 On occasion, the AMA will merge, delete, or introduce CPT I codes. However, the consistent structure of CPT I codes means that the mapping of a procedure to a group of related codes is generally stable. 21 For an example of typical reporting requirements, see Chapter 4 of the CMS Medicare Claims Processing Manual (CMS Title 100-04).

approve these requests. For rarer procedures, the process of requesting reimbursement may be costly and

payers may be slower to approve the payment. Importantly, providers can only report these unlisted codes

when the relevant procedures are not covered by a specific existing CPT code. For example, CMS instructs

Medicare contractors to verify that procedures cannot be covered by specific codes and to change submitted

claims accordingly. This implies that once a specific code is available, reporting is accurate—every provider

that reports the procedure must do so through the specific code. This is true even if it would be financially

advantageous to use a different existing code for the procedure, e.g., a provider cannot use an existing CPT I

code (reimbursable) to report when a newly issued CPT III code (oftentimes non-reimbursable) is available.22

Any stakeholder – the individuals who developed the new procedure, a medical society, or a device manufacturer – may apply for a new CPT code.23 For example, the Society for Cardiovascular Angiography

and Interventions, along with three other provider organizations, jointly applied for a set of CPT III codes

for TAVR in November 2009.24 An ad hoc review committee consisting of clinical and research specialists

meets two or three times every year to consider all applications. In considering whether to grant a new CPT

III code, the AMA committee considers the extent to which the new procedure differs from existing

procedures, the extent to which it is already in use, and any clinical evidence of effectiveness. In this process,

extra weight is given to peer-reviewed research supporting clinical efficacy. After approval, it can take one to two years before the CPT code is available for use. In the case of TAVR, the set of approved CPT III codes was not available until May 2012. After five years in CPT III status, the committee automatically sunsets

codes, unless there is an application for extension or promotion to CPT I. An applicant may request an early

promotion from CPT III to CPT I. For example, TAVR was promoted to CPT I status in January 2013.

2.4. CPT Codes and Procedure Utilization

One of the central questions in this paper is whether and how AMA coding decisions affect

procedure utilization, which is an important input to the investment decision of innovative firms. A few case

22 See Section 2.4 for detailed discussion on reimbursement for CPT I and CPT III codes. 23 If an applicant requests a CPT I code, the AMA may instead grant a CPT III code. 24 The set includes four codes pertaining to variations in the approach to the procedure and whether there was cardiopulmonary bypass.

studies show that promotions do lead to increased utilization for specific medical services, but there has been

no systematic analysis.25 To appreciate why the answer to this question is not obvious, it is important to be

clear about the implications of CPT I and CPT III codes. Neither code type implies explicit endorsement of a procedure by the AMA and neither code type assures reimbursement by payers.26 Treatment and

reimbursement decisions remain the independent responsibility of providers and payers. Furthermore, a CPT

III code does not imply that the AMA believes a procedure should eventually be assigned a CPT I code. On the other hand, a CPT III code also does not imply that the AMA believes a procedure is ‘experimental’. As a practical matter, all procedures must spend some time at CPT III before promotion to CPT I.

If physicians believe that a procedure is efficacious, the actual coding status could be theoretically irrelevant. In practice, coding matters for several reasons related to both reimbursement and information about treatment efficacy. First, most payers are reluctant to reimburse for CPT III codes. Indeed, many maintain blanket denials of reimbursement of all new CPT III codes. This is typically on the grounds that they represent ‘experimental’ or ‘medically unnecessary’ procedures, despite the AMA’s agnosticism on this point. Second, while CPT codes are not AMA endorsements, a successful AMA review, as reflected in the granting of a CPT I code, could be a positive signal of the procedure’s efficacy. In other words, although the

AMA is clear that CPT code assignments do not represent medical judgments (endorsements or otherwise), it is possible that some providers or payers interpret them this way.

3. Basic Facts about CPT Applications and Approvals

3.1 Extent of Medical Procedure Innovation

Our first goal is to document the extent of medical procedure innovation using a unique dataset on

all CPT code applications filed with the AMA between 2008 and 2017.27 Applications follow a fairly standard format and contain: (1) the identity of the applicant; (2) whether there is an associated device and whether the

25 See, e.g., Duszak et al. (2010) and Cox et al. (2016), for case studies. 26 As of 2019, each release of new CPT III codes contains the following text: “As with CPT I codes, inclusion of a descriptor and its associated code number does not represent endorsement by the AMA of any particular diagnostic or therapeutic procedure or service. Inclusion or exclusion of a procedure or service does not imply any health insurance coverage or reimbursement policy.” 27 We only observe the total count of applications filed in 2018 and 2019; no detailed information about these applications are provided to us. Applications filed before 2008 are only available in hard copy and were not produced for this study.

device has FDA approval; (3) published research findings pertaining to the procedure; (4) estimates of the

frequency of use and how many different providers use it; (5) identity of any medical societies that support

the application; (6) details about the procedure; and (7) how providers currently report the procedure for

reimbursement purposes.

Unfortunately, we do not have information about procedures that were developed or attempted to be

developed, but not ultimately proposed for AMA review. This could include procedures that failed early in

the process and those that could be sufficiently profitable under an existing code. Thus, unlike drugs, where

we have good early-stage research data and can estimate attrition rates from early in the process, we are

unable to do the same for procedures.

Table 1 summarizes information about CPT III applications between 2008 and 2019. Columns (1)-

(3) report numbers of applications, acceptances, and rejections by year. Columns (4)-(6) report the fate of

each accepted application, i.e., whether they are promoted to CPT I codes, sunsetted, or remain as CPT III

codes. The annual number of CPT III applications varies from 11 to 26, with a twelve-year total of 187. Of

these, 162 were approved. Of the 86 applications approved between 2008 and 2014, only 28 were promoted

to CPT I by 2019, while 32 remained as CPT III and 26 were sunsetted.28 Among the 28 promoted procedures, 25 were promoted within five years after the CPT III approval, which indicates the 5-year promotion rate is 29% (25/86).

Table 2 reports the statistics by procedure type. Panel 1 breaks down the statistics by whether the

application involves medical devices, and whether that medical device has been previously used for a different

procedure or approved more than two years prior to the CPT III application (“old” versus “new” device).29

Panel 2 reports results by type of applicant. There are slightly more industry applicants than medical society/physician applicants. Industry applicants have a slightly lower approval rate and a much lower conditional rate of promotion to CPT I codes.

28 We focus on the promotion and sunset of CPT III applications approved between 2008 and 2014 because the CPT III codes are valid for a maximum of five years if not extended. 29 We observe procedure characteristics such as device type, applicant type, and the innovation timeline for the 128 CPT III procedures approved between 2008 and 2017.

3.2 Development and Administrative Approval Times of Medical Procedures

Numerous studies estimate and discuss the average innovation timeline for pharmaceuticals. In this

section, we aim to develop a roughly comparable timeline for medical procedures that accounts for the

unique features of innovation in this area. We show that a meaningful portion of the development time for

medical procedures is the time period between the FDA’s approval of the associated device and the AMA’s

awarding of a permanent reimbursable billing code (CPT I code). Accounting for this administrative process,

we estimate that these procedures can face a longer development process than pharmaceutical products.

Measuring the innovation timeline is not straightforward, as it requires defining and identifying

beginning and endpoints. Studies of drug development use a variety of starting points, such as the first clinical

trials, and usually choose the date of FDA approval as the endpoint.30 For example, DiMasi et al. (2016) report that in 1990-2010, clinical trials for drugs require 95.2 months, on average. It takes another 16 months for FDA approval, for an average total development time of 9.2 years. They also report that the average time from the synthesis of a new drug to clinical trials is 31.2 months, giving a total time from the synthesis of a new drug to its approval of nearly 12 years. In contrast, Stern (2017) finds that the total development time for

Class III medical devices is 54 months, on average, using the onset of clinical trials as the starting point and

FDA approval as the endpoint.

With our unique access to CPT application data, we are able to extend the Stern (2017) time window in both directions, as well as measure development times for procedures that do not involve newly patented devices. As mentioned previously, there is no universally accepted definition of the starting point of innovation, which makes it difficult to compare development times across products and processes. With this in mind, we define the beginning of procedure innovation to be the first utilization of the procedure among humans, as reported in the first published research and/or the AMA applications.31 Although some procedures benefit from earlier animal model studies or lab studies, we use the first in-human utilization

30 See, e.g., Dranove & Meltzer (1994), Keyhani et al. (2006), DiMasi & Grabowski (2007), and DiMasi et al. (2016). 31 For example, the initial TAVR procedures started in 1985 but were unsuccessful due to the limitation of the available devices (Cribier et al. 1986). In 1999, Edward Lifesciences invented a new bioprosthetic heart valve sutured onto a balloon expandable stent. Using this device, the first in-human percutaneous aortic valve implantation was successfully performed in April 2002 as a “last resort” treatment for a patient in France (Cribier et al. 2002). Phase I pilot trial started in August 2003 (Cribier et al. 2006). In this case, we use the first in-human utilization of the procedure (i.e., April 2002) as the starting point when calculating the time span.

when calculating the time span because it is reported in the vast majority of CPT III applications. This is

roughly comparable to the start of phase I in drug trials. For procedures that involve FDA-approved/cleared

devices, the second milestone is the FDA approval or clearance of the device.32 The third milestone is the

CPT III approval by the AMA. Some of the CPT III procedures are eventually promoted to CPT I, while

others are sunsetted or temporarily extended by the AMA after five years from the initial CPT III code

publication.33

Table 3 presents summary evidence on the innovation timeline up through the effective date of the

CPT III approval. We report these estimates separately for innovations involving new devices, old devices,

and no devices. The first two rows include information available for all CPT III applications. The last three

rows contain information for the 75 CPT III applications involving devices for which we have information

about the associated FDA application.

The development time of new procedures is quite long – 134 months (11.2 years) on average

between the initial research study and CPT III approval. Most of this time occurs prior to the CPT III

application. The average lag between the CPT III application and the effective date for the new CPT III code

is only 14 months. This shows a relatively swift administrative process. That said, stakeholders of innovative

medical procedures including the device companies face a long step, 52 months (4.3 years), after FDA

approval before the associated procedure gets a provisional billing code (the CPT III code) with no guarantee

for reimbursement. By comparison, innovative pharmaceutical firms generally do not face such a long lag

after FDA approval before they can be reimbursed for their new product. In addition, there is considerable

variation in the time to CPT III approval across procedure types. The longest development times are for old

devices, perhaps because providers can usually bill for the new procedures using old codes and are therefore

in no rush to secure new codes. Figure 1 shows the distribution of the development time for all observed

CPT III approvals, ranging from 1 to more than 20 years.

32 We use the approval time of the 510K containing the intended use of the device as described in the CPT III procedure, rather than that of the first 510K clearance of the device. 33 The average regulatory review period from the submission of a new CPT III application (or an application for promotion from CPT III to CPT I code) to the effective date is about one year. For example, applications submitted in July 2017 are discussed in September 2017 meeting, and, if approved, become effective on July 1, 2018.

Table 4 repeats the analysis but restricts the sample to the 30 procedures promoted to CPT I by

2019. It takes an average of 194 months (16.1 years) from the first research study until the effective date of the CPT I code. This puts the development time for procedures on a par with, or even longer than, new drugs. Nearly half of this time is spent before the FDA submission. There are another 48 months (4 years) between FDA submission and CPT III application, although this is largely for procedures involving old devices, i.e., those that had previously been approved for another use. Finally, a full 38 months (3.2 years) is

spent between CPT III approval and the effective CPT I date. There is considerable variation in innovation

time: Appendix Figure A1 shows the distribution of promotion time from CPT III approval to CPT I

promotion and Appendix Figure A2 shows the distribution of overall time from first research to CPT I code

approval.

4. Effect of CPT Promotion on Utilization

4.1. Empirical Strategy

When developing innovative medical procedures or products, it is important to understand how the

procedure or product will be adopted and paid for in the marketplace. In this section, we show the

promotion of procedure billing codes from CPT III (provisional, non-reimbursable) to CPT I (provisional,

reimbursable) has a significant impact on utilization, which indicates that it must be considered in any analysis

of the appropriability of investments in procedural innovation.

Ideally, in order to study the effect of administrative coding decision on utilization, we would have

data on every time a medical provider uses the procedure regardless of the coding. Recall that providers are

required to use appropriate CPT III codes once they are available, which means that procedures should not

be occurring under other codes after the issuing of a CPT III code. Using AMA documentation, we are able

to match CPT III to CPT I codes for new procedures.34 This allows us to examine the effect on use from

34 The other possible approach is to find overlapping claims using different procedure code sets. There is one candidate for this in the International Statistical Classification of Diseases and Related Health Problems (ICD) code set. Unfortunately, ICD procedure codes are typically reported in inpatient settings so do not track use of outpatient procedures. In addition, during the period covered by our data the relevant version of the ICD code set was the ICD-9 set. The ICD-9 code set is generally less detailed than the CPT code set,

granting a CPT I code. Unfortunately, we cannot identify procedure use prior to the assignment of the CPT

III code and therefore cannot empirically estimate the effect of being granted a CPT III code.

The second empirical challenge is identification. First, since prior utilization is a criterion for

procedure promotion and is also relevant when providers make decisions about use, we are concerned that

promoted procedures are systematically different from non-promoted procedures. Therefore, we rely on

“own case control” (i.e., procedure-specific fixed effects) to measure the bump in utilization for each

promoted procedure. Also, to address the concern that the increase in procedure utilization after promotion

may reflect a continuation of the increasing trend in the pre-promotion period, it is crucial to examine

whether the utilization increased discontinuously at the time when CPT codes are promoted. We do so using

an event study approach.

Specifically, we employ the following two specifications to estimate the effect of CPT code

promotion on procedure utilization.

= + + + + (1)

𝑌𝑌𝑖𝑖𝑖𝑖 𝛽𝛽0 𝛽𝛽1𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝐼𝐼𝑖𝑖𝑖𝑖 𝜃𝜃𝑋𝑋𝑖𝑖𝑖𝑖 Υ𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖 Θ𝑌𝑌𝑌𝑌𝑌𝑌𝑟𝑟𝑡𝑡 = + _ ( ) + + + (2)

𝑖𝑖𝑖𝑖 0 𝑑𝑑 𝑖𝑖𝑖𝑖 𝑡𝑡 𝑖𝑖𝑖𝑖 𝑖𝑖 𝑡𝑡 The first𝑌𝑌 specification𝛽𝛽 Φ, ∑Equ𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶ation (1),𝐸𝐸𝐸𝐸 𝐸𝐸is𝐸𝐸 a𝐸𝐸 time-varying𝜃𝜃𝑋𝑋 differenceΥ𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃-in-differenceΘ𝑌𝑌𝑌𝑌s𝑌𝑌 (DID)𝑟𝑟 estimation.

The dependent variable takes two forms: a continuous variable representing the utilization (i.e., the

𝑖𝑖𝑖𝑖 natural log of the number𝑌𝑌 of Medicare services) of procedure in year , and an indicator variable for whether procedure records any Medicare utilization in year . 𝑖𝑖 is an𝑡𝑡 indicator variable which equals 1 if

𝑖𝑖𝑖𝑖 procedure 𝑖𝑖 has been assigned a CPT I code in year 𝑡𝑡. 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 represents𝐼𝐼 time-varying procedure characteristics—

𝑖𝑖𝑖𝑖 we include 𝑖𝑖a categorical variable for whether procedure𝑡𝑡 𝑋𝑋 involves device and whether the associated device

has been approved by the FDA by year . In a robustness𝑖𝑖 test, we also control for the interaction between the number of years the procedure code has𝑡𝑡 been effective since its initial CPT III approval, , and the

𝑖𝑖𝑖𝑖 log time trend to allow procedures in different tenure stages to have diff𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇erent trajectories of

𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝑡𝑡

so new procedures that receive new CPT codes may not receive new ICD-9 codes. The recent adoption of the more detailed ICD-10 code set means that this approach might be more feasible in the future.

utilization growth. The coefficient of interest is . If promotion from CPT III to CPT I has a positive

1 impact on utilization, we expect to be positive.𝛽𝛽

A key assumption of validating𝛽𝛽1 the DID estimation is the parallel trend assumption, i.e., promoted

procedures have similar utilization trends as non-promoted procedures in the pre-promotion period. To

better assess this assumption and also to examine the dynamic effects, we estimate an event study model,

Equation 2. Specifically, we replace the post-time dummy with a set of dummy variables

𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝐼𝐼𝑖𝑖𝑖𝑖 _ ( ), indicating both leads and lags from year relative to the year of code promotion.

𝑑𝑑 𝑖𝑖𝑖𝑖 𝑡𝑡 ∑ 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝑡𝑡

4.2 Data

The main data source for procedure utilization is the CMS Medicare Provider Utilization and

Payment Data in 2012-2017. The data provides annual CPT code-level procedure utilization. It covers all procedures provided to Medicare beneficiaries enrolled in Medicare part B and includes both inpatient and outpatient procedures. The utilization of a procedure in a certain year is measured by the number of Medicare services reported in this data. We supplement the utilization data with information on procedure promotion date and procedure characteristics obtained from the AMA CPT code documentation. We also use the AMA documentation to aggregate related CPT codes that represent the same procedure and to match CPT III and

CPT I codes for promoted procedures.35 We extract utilization data for all CPT III codes created since 2001

that remained active in 2012-2017, as well as those that were promoted to CPT I codes.

We identify a total of 801 procedure-year observations, representing 184 procedures with active CPT

III codes between 2012 and 2017 (Sample 1). Of the 184 procedures, only 72 recorded Medicare utilization in

some years between 2012 and 2017, representing a subsample of 319 procedure-year observations (Sample 2).

Furthermore, among the 319 procedure-year observations, only 240 observations record Medicare utilization in the given year (Sample 3). This relatively low rate of Medicare utilization record likely reflects the fact that

35 CPT codes are primarily administrative, so a single procedure may be assigned to a range of codes. This allows providers to report common procedure variations which might involve different costs. Since the underlying technology and techniques are the same across these codes, it is more appropriate to group them as a single procedure for our analysis. There is also often not a one-to-one mapping from CPT III codes to CPT I codes after promotion.

procedures with utilization of fewer than 11 cases are unreported in the CMS data. Also, not all new

procedures are relevant for the patient population covered by Medicare (i.e. the elderly and disabled). Among

the 72 CPT III procedures with Medicare utilization, 24 percent (17 procedures) were promoted to CPT I

between 2012 and 2017; none of the 112 CPT III procedures with unreported Medicare utilization were promoted during the study period.

Since it is unclear whether the unreported utilization indicates zero utilization or missing positive values, we use several alternative ways to handle the unreported utilization in the CMS data. First, our preferred method is to fill the unreported value in Samples 1 and 2 with 10, which leads to a lower bound estimate of the billing code promotion effect due to the fact that pre-promotion observations are more likely to have unreported utilization. Second, for robustness, we also consider filling the unreported values with 1,

0, and a randomly drawn integer between 1 and 10. Finally, in Sample 3, we drop the unreported values.

Table 5 presents the summary statistics of the three samples.36

4.3 Results

The main results of DID estimation from Equation (1) are shown in Table 6. Our most preferred

specification is based on the full sample with unreported utilization replaced by 10, which generates a lower

bound estimate of the administrative coding effect. We find that CPT code promotion from Category III to

Category I is associated with a 9-fold increase in utilization (Table 6, Column 1).37 As expected, the effect

becomes even larger when we focus on the subsample of procedures with recorded Medicare utilization

(Table 6, Column 4), or fill the unreported utilization with 1 instead of 10 (Table 6, Columns 2 and 5). Next,

when treating the unreported value as no utilization, we find that CPT code promotion is associated with a

31.9% higher odds of any utilization (Table 6, Column 3), and this effect again becomes larger when

restricting to the subsample of procedures with recorded Medicare utilization (Table 6, Column 6). Finally,

36 Note that the number of observations with procedure characteristics (i.e., whether involving exclusive or nonexclusive devices or applied by medical societies) is less than the total because we only observe procedure characteristics for those approved after 2007. 37 . 1 = 9.26 2 328 𝑒𝑒 − 22

the result remains highly robust when we drop all observations with unreported utilization (Table 6, Column

7).

To validate the parallel assumption of the DID estimation and examine the dynamic effect, we

present in Figure 2 the estimated coefficients from Equation (2).38 Utilization increased discontinuously

starting from the promotion year and it stays relatively stable throughout the post-promotion period. There is

no anticipatory effect in the years prior to promotion. This finding suggests that there is a one-time increase in utilization upon promotion to Category I codes, but no significant change in the diffusion rate afterward.

These estimates demonstrate that the administrative coding decision plays an important role in the ultimate use and diffusion of that procedure.

We offer one caveat to this analysis. An important limitation of this analysis is that there may be innovative procedures for which providers are content to bill under existing codes. Our data do not allow us to identify these procedures. Thus, we are likely overstating the extent to which obtaining a new CPT I code impedes utilization of all new procedures. In Section 5.4, we discuss in detail the new procedures reported in old codes and explain why the overstatement of our analysis is likely to be small. For those procedures that are not well-accommodated by existing codes, however, it is clear that coding matters.

4.4 Extensions—Heterogeneity and Robustness

In this section, we first test the heterogeneity of the main effect by procedure type, i.e., whether the procedure involves exclusive patented devices, nonexclusive devices, or is applied by medical societies as opposed to industry firms (Appendix Table A1). Appendix Table A1, Column 2, suggests that the increase in utilization due to CPT code promotion is stronger among procedures that involve medical devices (both exclusive and nonexclusive devices). There could be several explanations: device manufacturers might invest more in promoting their devices when the relevant procedures are reimbursed; procedures involving devices

38 Figure 2 is based on the full sample with unreported utilization replaced by 10. Results remain highly robust when using the subsample of procedures with recorded Medicare utilization (Appendix Figure A3).

might be difficult to claim under existing codes and procedures without devices might allow more flexibility

in claiming.

Next, we perform a number of tests to show our main findings are robust to the selection of model specification and sample. First, as mentioned above, the CMS data does not report utilization for procedures with 10 or fewer cases in the year. In our baseline tests, we assigned the value of 10, 1, and 0 to unreported

utilization. However, the missing value for these procedures could be any number between 1 and 10.

Therefore, we re-run the analysis (Equation 1) 10,000 times with different random draws for the missing

values of utilization from a discrete uniform {1, 10} distribution. Appendix Figure A4 shows the results of

the estimated coefficients and standard errors of the post-CPT I dummy variable in Equation 1. Appendix

Figure A2, Panel (a) shows the results when using the full sample, and panel (b) shows the results using the

subsample of procedures with recorded Medicare utilization. The results are highly robust and remain

statistically significant.

The second robustness test re-estimates Equation (1) by including the interaction term between the tenure of

the procedure (i.e., number of years since gaining a unique CPT III code) and the log time trend. This

specification captures potential differential trends in diffusion rates across procedures in different tenure

stages. The third robustness test aims to exclude the potential effect of FDA device approval on utilization.

Although we controlled for the time-varying variable in the main specification on whether the procedure

involves a device that hasn’t been approved by the FDA at the time of gaining a unique CPT III code, we

conduct a robustness test by re-estimating Equation (1) using a subsample that excludes all procedures that

involve unapproved devices at the time of CPT III approval. The results from these two tests are shown in

Appendix Table A2 and A3, respectively. The results remain highly robust and quantitatively similar to the

main results shown in Table 6. Last, since we use code publication date as the date of promotion in the main analysis, yet the promotion decision is normally made during the year prior to the year in which the code is published, we perform a robustness test by setting the promotion year based on the year prior to the publication year. We re-estimate Equation (1) and show the robust results in Appendix Table A4. Compared with the main analysis shown in Table 6, the effects become smaller but remain highly significant.

4.5 Does CPT III Status Affect Utilization?

While we have documented that promotion to CPT I status has a profound impact on utilization, we cannot readily quantify the effect of the AMA granting of CPT III status, as there is no consistent documentation of utilization for procedures that do not have their own CPT codes. Several qualitative factors suggest that insurers generally do not reimburse for procedures with CPT III status, which in turn suggests that utilization may be limited. Perhaps most importantly, CMS does not assign RVUs to CPT III codes, which means that they cannot be billed like other procedures.39 Indeed, insurer contracts with providers often

include a blanket denial of reimbursement for Category III procedures, making a small number of exceptions

for specific treatments. For example, the March 2019 Policy Guideline for UnitedHealthcare’s Medicare

Advantage program, notes that, except under specific circumstances, “UnitedHealthcare considers all services

and procedures listed in the current and future CPT III code list as not proven effective and will deny

submitted claims as not medically necessary.”40 In the accompanying list of CPT III codes, more than 85 percent are listed as noncovered, and all but one of the covered codes are covered with restrictions.41

Documents from medical device manufacturers provide additional evidence that reimbursement of

CPT III codes can be difficult. Respiri (formerly iSonea) develops asthma monitoring devices. Its 2011

Annual Report describes receiving a Category III code for one of its products as a key achievement of the

preceding year.42 However, it also notes that upgrading the code to CPT I is necessary for securing reimbursement in the US market: “Achievement of a Category I CPT code will enable the company’s products to become reimbursed by CMS and the majority of US health insurance payers.” To take another example, Si-Bone develops minimally invasive surgical treatments for sacroiliac joint disorders. It completed

39 Source: AMA CPT Category III Codes Long Descriptors https://www.ama-assn.org/system/files/2020-07/cpt-category3-codes- long-descriptors.pdf Retrieved on 8/23/ 2020. 40 Source: UnitedHealthCare Medicare Advantage Policy Guideline Category III CPT Codes. https://www.uhcprovider.com/content/dam/provider/docs/public/policies/medadv-guidelines/c/category-iii-cpt-codes.pdf Retrieved on 8/23/2020. 41 For example, coverage is only granted for certain indications, e.g., meeting the FDA-approved protocols for IDE clinical trials, or performed under coverage with evidence development (CED) when a clinical study meets certain criteria. 42 Source: iSonea Annual Financial Report. https://www.asx.com.au/asxpdf/20111019/pdf/421vl9t3rc2xfq.pdf Retrieved on 8/23/2020.

an initial public offering (IPO) in 2018 and included a detailed discussion of the relationship between CPT

codes and reimbursement in its prospectus.43 According to this document, the creation of a new CPT III code for the procedure involving their product in 2013 may have slowed adoption. The reason is that the minimally invasive procedure was previously claimed under the CPT I codes for the invasive version of the procedure. The new CPT III code threatened reimbursement because, as the prospectus notes, CPT III codes are reimbursed “sporadically”. Positive coverage decisions by payers were delayed until after a new CPT I code became effective in 2015.

5. Contrasting Drug and Procedure Innovation

5.1 The Pace of Innovation

In documenting the CPT approval process, one may be struck by the relatively small numbers of new

procedures. Over the past ten years, the AMA has approved an average of fewer than 14 procedures for CPT

III status, and only about 4.3 per year for CPT I promotion. In contrast, the FDA has approved an average of

44 new chemical entities and therapeutic biological products each year during the most recent five years.

We show Medicare utilization of the top 5 utilized new medical procedures and new drugs in Appendix

Tables A5 and A6, respectively, where new procedures and drugs are defined as those approved within five years of the corresponding year. The comparison between the two tables suggests that utilization is higher for top-utilized new drugs than top-utilized new procedures. Looking at the data another way, Appendix Table

A7 shows that total Medicare spending on new procedures during 2013-17 was only 21 percent of spending on new drugs.44 This results from both the small number of identifiable new procedures compared to new drugs and differential pricing across the two categories.

Next, we show in Appendix Tables A8 and A9 the year of introduction of the top 15 utilized medical procedures and top 15 utilized drugs by Medicare beneficiaries in 2017. Examining these data, one may be tempted to conclude that the pace of drug innovation has outstripped the pace of procedure innovation.

43 Source: SiBone IPO Report. https://investor.si-bone.com/static-files/83477437-1762-4a30-be4a-43a8c88968eb Retrieved on 8/23/2020. 44 Because we no longer count TAVR as a new procedure in 2016, there is sharp drop in new procedure payments that year.

However, we caution against such a stark interpretation. This pattern could also reflect the relatively

unmonitored development process for procedures. Each stage of every new drug (even those involving incremental improvements to an existing product such as a change in the delivery mechanism) is meticulously tracked by the FDA. Therefore, even small changes to pharmaceutical products are readily apparent in the data. In contrast, many (if not most) procedures are continually improved without any formal review and therefore are unobservable in any systematic fashion. As a result, the data on incremental advancements in new procedures is almost certainly under-reported in the data.

That said, there are also reasons to be concerned that the observed differences in the rate of innovation across the two categories are real, i.e. that we are experiencing more product than procedural innovation. These concerns stem from the institutions surrounding the innovative process. In particular, the lack of comparable property rights across the two categories could create differences in the pace of innovation for procedures compared to drugs.

As we discuss below, the deliberative process of regulatory oversight for drugs has created well-

established rules for product development as well as clear and strong property rights. These rules represent a

considered tradeoff among three factors: safety, innovative incentives, and monopoly pricing.

In contrast, the ad hoc process for oversight regarding innovation for medical procedures creates

uncertainty over both the development process and the allocation and enforceability of property rights.

There is little evidence of a considered tradeoff in policies that address the ways in which a lack of

appropriability could cause potential underinvestment in value-creating procedures. As a result, the existing

pace of process innovation could only be optimal by chance.

5.2 Contrasting Rules and Regulations Regarding Innovation

The Federal Food, Drug, and Cosmetic Act of 1938 gave the FDA authority to oversee drug safety.

After concerns about the safety of approved medications, notably thalidomide, the FDA Amendments of

1962 codified the testing requirements for drugs. The resulting structure for developing drugs – from preclinical trials through three phases of Investigational New Drug trials and final FDA review, all supervised

by the FDA – remains essentially unchanged. Once the FDA approves a drug, insurers nearly always agree to

pay for it, although the price may be subject to negotiation. Over time, the FDA has taken steps to accelerate

approval of “important drugs”.45 The FDA continues to modify the review process, as witnessed by recent

efforts to accelerate the review of COVID-19 treatments and vaccines. Through the Drug Price Competition

and Patent Term Restoration Act (Hatch-Waxman Act) of 1984, Congress increased the effective patent lives of drugs that required lengthy reviews, while facilitating entry by generics once patents expired. These changes suggest that regulators are carefully weighing safety, development times, and protection of property rights. While stakeholders debate whether patent lives should be extended and whether testing is too rigorous, few take meaningful issue with the general structure of the drug approval process.46

In contrast, there is minimal regulatory oversight of medical procedures. Any medical provider can

perform any medical procedure that they deem necessary and appropriate, even a procedure that no one has

previously performed. There are two legal obstacles. The first is the malpractice risk. The second arises with

procedures involving new medical devices. The FDA must approve a device for systematic clinical use in a

specific procedure, a process that requires much less formal testing and development time than the process

for drugs.47 Once the FDA approves a device for one procedure, providers may use it for other procedures

without additional approval. That said, the FDA does not ban inventive activities of physicians that lead to

new medical devices, or oftentimes, new patents. In fact, physician-generated patents play a more important

role in subsequent R&D activities by device manufacturers than non-physician patents (Chatterji et al. 2008;

Smith & Sfekas 2013).

Without AMA approval of a CPT I code, however, providers cannot be assured of reimbursement,

even if the procedure involves an FDA-approved device. The centrality of CPT coding and the central role played by the AMA seems almost accidental. The AMA initially classified diseases in the 1872 Nomenclature of

45 See Dranove and Meltzer (1994). 46 For details about the Federal Food, Drug, and Cosmetic Act of 1938, see https://www.fda.gov/about-fda/fdas-evolving- regulatory-powers/part-ii-1938-food-drug-cosmetic-act. For details about the FDA Amendments of 1962, see https://www.fda.gov/consumers/consumer-updates/kefauver-harris-amendments-revolutionized-drug-development. For details about the Drug Price Competition and Patent Term Restoration Act of 1984, see https://www.fda.gov/drugs/abbreviated-new-drug- application-anda/hatch-waxman-letters. 47 See Stern (2017).

Diseases. In the 1942 edition of the Nomenclature, the AMA added a list of commonly performed operations. By

the 1960s, the AMA was expanding this list to include diagnostic tests and a range of medical treatments, and

in 1966 it published a copyrighted report entitled Current Procedural Terminology. CPT coding did not become central to reimbursement until 1983 when CMS sought to better codify Medicare payments. With no better alternative, CMS adopted CPT codes, and private insurers quickly followed suit. In this way, a coding system developed to facilitate the exchange of clinical information became the foundation for billing of medical procedures.

Whereas the FDA spells out testing requirements in great detail, the AMA offers little specific

guidance for innovators. The CPT III/CPT I promotion ladder resembles, at least superficially, the

Investigational New Drug (IND) development process for drugs. In addition, innovators that wish to fast

track approval may seek promotion to CPT I status prior to the expiration of the 5 year probation period.

There, the similarities end. There are no distinctions among stages comparable to the safety/small scale trials,

large trials distinctions of IND I, II, and III. The AMA does not specify sample sizes, type I, and type II

error criteria. Instead, applicants for Category III CPT codes must meet at least one of the three criteria: a)

support by at least one CPT/HCPAC advisor representing practitioners who would use the procedure, b)

support by peer-reviewed literature, and c) an IRB approved protocol of the clinical study, an ongoing U.S.

clinical trial, or other evidence of evolving clinical utilization.48 As a result, there is a wide range of research

methods used to assess the efficacy of new procedures. Among the 128 procedures approved by the AMA

for Category III CPT codes between 2008 and 2017, only 20 percent are supported by completed

Randomized Controlled Trials (RCTs). In contrast, Hatswell et al. (2016) document that 94 percent of the

774 drug approvals issued by the FDA between 1999 and 2014 are based on RCTs. Instead, 65 percent of

procedure applications are supported by non-Randomized Controlled Trials (RCT), and 15 percent are

supported by ongoing RCTs of which results are still pending at the time of the CPT III application.

48 Source: Criteria for CPT® Category I and Category III codes, https://www.ama-assn.org/practice-management/cpt/criteria-cpt- category-i-and-category-iii-codes Accessed on 8/23/2020.

Admittedly, it may be difficult to perform RCTs with many procedures, but the heterogeneity of evidence presented in support of CPT applications is striking.

5.3 Property Rights

Perhaps the biggest difference between drug and procedure development that could contribute to the pace of innovation involves property rights. While device makers hold patents, these patents seem to be easy

to invent around.49 For procedures, there are significant obstacles to obtaining and enforcing patents. In

1994, the AMA House of Delegates voted to condemn patenting of medical procedures. In 1998, the AMA

Council on Ethical and Judicial Affairs appealed to “the open exchange of information without the

expectation of financial reward for advancing medical science” (American Medical Association 1998). The

AMA still subscribes to this view and has pushed for legislation banning procedure patents (Anderson 1999).

It is also unclear whether courts will uphold procedure patents independent of the patent for the associated

medical device (Meier 1997). Perhaps for this reason, procedure patents are rarely enforced (Anderson 1999).

Another key consideration is reimbursement. Consider that the price of any product or procedure is

a function of the value that it creates. When a physician prescribes a patented drug, most of the value

captured by the producer goes to the drug company. In contrast, providers receive a substantial share of the

value created by a medical procedure. Thus, the financial benefits of a new procedure are diffused across both

the device maker (if there is a device) and potentially thousands of independent stakeholders. Facing a

commons problem, organized medicine has responded by initiating the CPT coding applications for new

procedures that lack property rights protection.

To shed some light on the role of property rights in the development process of medical procedures,

we examine heterogeneity in the applicants across procedures based on whether the procedure involves a

device and whether the device has non-expired patents. As previously noted, medical societies and physicians

represent nearly half of all applicants for new CPT III codes, and 84 percent of the applicants for successful

CPT I promotions. However, the data reveals systematic patterns in these applications. Table 7 reports the

49 As Goldman and Lakdawalla (2011) noted, the innovation process for medical devices is typically much faster than that for drugs.

type of CPT III applicant (industry versus medical society) based on whether the procedure involves an

exclusive patented device, nonexclusive device (e.g., off-patent device), or no device. Consistent with the aforementioned role of medical societies in solving the commons problem when there is a lack of property rights protection, we find that the vast majority (37 out of 42) of CPT III procedures with no device or off- patent devices are applied by medical societies, with the rest applied by consulting firms that are in the business of assisting physician groups with reimbursement issues (Table 7, Col 3 and 4). In contrast, only 21 out of 86 (24%) of CPT III procedures involving exclusive patented devices are applied by medical societies

(Table 7, Col 2).

5.4 New Procedures in Old Codes

The scope of our study focuses on new medical procedures that have entered the CPT system and gained unique CPT codes. Admittedly, there are also “new procedures in old codes (NPOCs)” i.e., new

procedures that are well-accommodated by existing codes and thus never assigned a unique CPT code. Our

data do not allow us to separately identify NPOCs from old procedures sharing the same codes.50 Therefore,

our analysis, which focuses on new CPT codes, likely understates the extent of new procedure innovation. In

this section, we conceptualize why some new procedures can be well-accommodated by old codes. We then

offer some reasons why, with a few caveats, any understatement of novel procedure innovation is likely to be

small.

As described in Section 2.2, there are two broad categories of new procedures based on the degree of

innovation.51 The first category is radically new procedures that are discontinuous from existing procedures

(Type I). Since these procedures share few common features with existing offerings, they can only be billed under “unlisted” CPT codes with no assigned RVUs before gaining unique CPT codes. Therefore, in order to get reimbursement, providers who perform these procedures have incentives to obtain new CPT codes. The second category is incrementally new procedures that are adapted from existing procedures and offer limited

50 The volume of old procedures may dwarf that of many NPOCs, making it difficult to detect any meaningful increase in the use of the shared codes. 51 See Schumpeter (1934) for definitions of radical and incremental innovations.

changes; without unique CPT codes, they may be billed under existing codes. Within this second category,

there are two different types: cost-increasing incrementally new procedures, i.e., those cost more than the existing

alternatives (Type II), and cost-decreasing incrementally new procedures, i.e., those cost less than the existing ones

(Type III). Similar to Type I procedures, billing Type II procedures under existing codes that were designed for low-cost alternatives reduces providers’ profits; therefore, providers have incentives to obtain new CPT codes with higher assigned RVUs. This does not apply to Type III procedures, however, because billing Type

III procedures under existing codes is profitable, and thus providers have little incentive to apply for new codes. This means Type III new procedures are the NPOCs that are well-accommodated under existing codes and are unobservable to researchers.

Next, we compare the relative prevalence of these three types of new procedures based on our data

and explain why Type III procedures are unlikely to be widespread, which reduces our concern about the

overstatement of the CPT I effect on the utilization of new procedures. First, recall that providers are

required to bill under the most appropriate codes when available, which means a procedure cannot be

reported under existing codes once a new CPT code is issued. This indicates that we observe the correct

utilization of Type I and Type II procedures conditional on new codes being issued. In our data, among the

128 procedures that have been issued new CPT III codes during the study period, 79 were reported under

unlisted codes with no assigned RVUs before obtaining new codes—the fact that there are no appropriate

existing codes for these procedures indicates that they are Type I procedures; 49 were reported under existing

codes with assigned RVUs, and they can be deemed as Type II procedures. This finding suggests that Type II

procedures are less prevalent than Type I procedures.

What about Type III procedures, which have the potential to reduce costs? Weisbrod (1991)

persuasively argued that the healthcare system historically favored cost-increasing (and quality-enhancing)

technologies rather than cost-decreasing technologies (Weisbrod 1991). If this still holds true, then Type III

procedures may be even less prevalent than Type II procedures.

A glaring exception to our taxonomy of procedures involves new technologies that had the potential

to improve outcomes at a higher cost but have failed to deliver. This is exemplified by robotic-assisted

surgery, in which a physician-guided robot is a substitute in production for a laparoscopic surgeon. Since the initial FDA approval of the da Vinci robot for clinical use in 2000, robotic-assisted procedures have been increasingly adopted for gynecologic, prostate, head and neck, and other surgeries. Accounting for both fixed and variable costs, including the costs of the robot and the costs of the surgeon’s time, robotic surgery is probably, but not definitively, more costly than hands-on surgery.52 There is also no clear evidence that it offers meaningfully superior outcomes (Wilensky 2016; Wright 2017). As a result, the AMA has determined that it is unnecessary to issue unique CPT codes for robotic-assisted procedures.53 Instead, the use of a robot

is considered integral to the performance of laparoscopic procedures and should be billed under existing

codes.54 The da Vinci robot is the culmination of large investments in procedure innovation, and the use of

robotic surgery continues to grow despite the mixed evidence on outcomes. Even so, a taxonomy of

procedure innovation based on CPT coding must inevitably miss investments like these, even if they translate

into commercially successful products.

We finally note that we do observe incremental innovations to existing procedures. These include

drugs that complement procedures (e.g., immunosuppressants for transplant surgery), diagnostics that

facilitate procedure improvements (e.g., 3T MRIs used in conjunction with prostate biopsies), better

prosthetics, and changes in the way that procedures are performed (e.g., off-pump open heart surgery).

Continuing improvements in outcomes for patients undergoing a wide range of procedures suggest that these incremental innovations may be equally or more important than the development of new medical procedures.

6. Conclusion

In this paper, we seek to explore inside the black box of the economics of medical procedure innovation and contrast it with the previously well-documented innovation process of pharmaceutical

52 The fixed cost of purchasing a da Vinci robot is about 2 million USD; the incremental cost ranges from 3,000 to 6,000 USD per patient (Wilensky 2016). 53 Source: Kaiser Permanente Clinical Review Criteria—Robotic Assisted Surgeries. https://provider.ghc.org/all- sites/clinical/criteria/pdf/robotic_assist_surgeries.pdf Accessed on 8/16/2020. 54 Note that physicians can report their use of robotic-assistant procedures by attaching an add-on HCPCS code (S2900) to the primary laparoscopic procedure; however, this add-on code is not reported in administrative databases such as the Medicare Provider Utilization and Payment Data used in our study. For inpatient facilities, there has been no unique DRG codes for providers to bill robotic surgery, although providers can use ICD-9 codes to report their utilization.

products. Using a novel and proprietary dataset on CPT applications of emerging medical procedures, we

unveiled, for the first time in the literature, the extent and overall timeline of an important subset of medical

procedure innovations - those receiving consideration for new billing codes. The ten-year estimate from initial research to obtaining unique procedure codes is much longer than previously reported for the innovation process of medical devices, and is as long or longer than that of new drugs; there are also meaningful variations in innovation times across procedures, depending on whether the procedure involves medical devices and the type of devices.

Compared with pharmaceutical innovations, we highlight two striking features of medical procedure innovations were highlighted. First, many new procedures, especially those that do not involve recently developed devices, lack property rights. This creates a potential appropriability issue and suboptimal innovation incentives. In many cases, physicians address this problem through their specialty medical societies, which are responsible for the majority of applications for billing codes. Second, administrative coding decisions can be crucial to the success of procedure innovations; we found that promotion to a CPT I code from a CPT III code has a large, positive effect on new procedure use. This finding is consistent with qualitative evidence from payers and medical device manufacturers and stands in stark contrast with drugs, where FDA approval is, with rare exceptions, sufficient to trigger reimbursements from payers.

There are several limitations to this study that may point ways to future research. First, as mentioned in Section 5.4, the scope of this study focuses on new medical procedures that have obtained unique administrative procedure codes; therefore, our findings do not speak to new procedures in old codes or incremental innovations to existing procedures. Future work is needed to assess the full picture of economic incentives behind all levels of inventive activities by firms and physicians that lead to new medical procedures.

Second, a better understanding of the clinical and economic value of procedure innovation is essential to evaluate the welfare consequences of current and alternative regulatory environments, as well as the coding and billing systems, regarding medical procedures. In particular, economists had long made a distinction between breakthrough and “me-too” drugs, and research effort has been devoted to examine their distinct values (see, e.g., Kessler et al. 1994; Lu and Comanor 1998; Towse and Leighton 1999; Werthheimer et al.

2001; Lee 2004; DiMasi and Paquette 2012); comparatively little is known about new medical procedures.

Finally, successful invention and adoption of innovative medical procedures involve many stakeholders, including patients, healthcare providers, professional medical societies, medical device companies, payers, as well as regulatory agencies and agencies that establish and maintain the billing and coding systems. Future work is needed to investigate the role of each stakeholder, the interplay among them, and the overall ecosystem surrounding procedure innovation.

References

Acemoglu, Daron, and Joshua Linn. 2004. "Market size in innovation: theory and evidence from the pharmaceutical industry." The Quarterly Journal of Economics 119 (3):1049-1090. Agha, Leila, Soomi Kim, and Danielle Li. 2020. "Insurance Design and Pharmaceutical Innovation." NBER Working Paper No. 27563. Agha, Leila, and David Molitor. 2018. "The local influence of pioneer investigators on technology adoption: evidence from new cancer drugs." Review of Economics and Statistics 100 (1):29-44. American Medical Association, Council on Ethical, and Judicial Affairs. 1998. "Ethical issues in the patenting of medical procedures." Food and Drug Law Journal:341-351. Anderson, Scott D. 1999. "A right without a remedy: the unenforceable medical procedure patent." Marquette Intellectual Property Law Review 3:117. Aouad, Marion, Timothy T Brown, and Christopher M Whaley. 2019. "Reference pricing: The case of screening colonoscopies." Journal of Health Economics 65:246-259. Blume-Kohout, Margaret E, and Neeraj Sood. 2013. "Market size and innovation: Effects of Medicare Part D on pharmaceutical research and development." Journal of Public Economics 97:327-336. Carroll, Caitlin, Michael Chernew, A Mark Fendrick, Joe Thompson, and Sherri Rose. 2018. "Effects of episode-based payment on health care spending and utilization: Evidence from perinatal care in Arkansas." Journal of Health Economics 61:47-62. Cox, Kelly, Richard Duszak, Jennifer Hemingway, Danny R Hughes, and Sadhna B Nandwana. 2016. "Reassessing medicare trends in diagnostic CT colonography after achieving CPT code category I status." Abdominal Radiology 41 (7):1357-1362. Cribier, Alain, Helene Eltchaninoff, Assaf Bash, Nicolas Borenstein, Christophe Tron, Fabrice Bauer, Genevieve Derumeaux, Frederic Anselme, François Laborde, and Martin B Leon. 2002. "Percutaneous transcatheter implantation of an aortic valve prosthesis for calcific aortic stenosis: first human case description." Circulation 106 (24):3006-3008. Cribier, Alain, Helene Eltchaninoff, Christophe Tron, Fabrice Bauer, Carla Agatiello, Deborah Nercolini, Sydney Tapiero, Pierre-Yves Litzler, Jean-Paul Bessou, and Vasilis Babaliaros. 2006. "Treatment of calcific aortic stenosis with the percutaneous heart valve: mid-term follow-up from the initial feasibility studies: the French experience." Journal of the American College of Cardiology 47 (6):1214-1223. Cribier, Alain, Nadir Saoudi, Jacques Berland, Thierry Savin, Paulo Rocha, and Brice Letac. 1986. "Percutaneous transluminal valvuloplasty of acquired aortic stenosis in elderly patients: an alternative to valve replacement?" The Lancet 327 (8472):63-67. Cutler, David M. 2004. Your money or your life: Strong medicine for America's health care system: Oxford University Press. Darrow, Jonathan J. 2017. "Explaining the absence of surgical procedure regulation." Cornell Journal of Law and Public Policy 27:189. DiMasi, Joseph A, and Henry G Grabowski. 2007. "The cost of biopharmaceutical R&D: is biotech different?" Managerial and Decision Economics 28 (4‐5):469-479. DiMasi, Joseph A, Henry G Grabowski, and Ronald W Hansen. 2016. "Innovation in the pharmaceutical industry: new estimates of R&D costs." Journal of Health Economics 47:20-33. DiMasi, Joseph A, Ronald W Hansen, and Henry G Grabowski. 2003. "The price of innovation: new estimates of drug development costs." Journal of Health Economics 22 (2):151-185. DiMasi, Joseph A, and Cherie Paquette. 2004. "The economics of follow-on drug research and development." Pharmacoeconomics 22 (2):1-14. Dranove, David, Craig Garthwaite, and Manuel I Hermosilla. 2020. "Expected Profits and The Scientific Novelty of Innovation." NBER Working Paper No. 27093. Dranove, David, and David Meltzer. 1994. "Do important drugs reach the market sooner?" The RAND Journal of Economics:402-423. Dubois, Pierre, Olivier De Mouzon, Fiona Scott‐Morton, and Paul Seabright. 2015. "Market size and pharmaceutical innovation." The RAND Journal of Economics 46 (4):844-871.

Duszak, Richard, Robert J Optican, Kenneth P Brin, and Pamela K Woodard. 2011. "Cardiac CT and coronary CTA: early Medicare claims analysis of national and regional utilization and coverage." Journal of the American College of Radiology 8 (8):549-555. Finkelstein, Amy. 2004. "Static and dynamic effects of health policy: Evidence from the vaccine industry." The Quarterly Journal of Economics 119 (2):527-564. Goetz, RH. 1961. "Internal mammary-coronary artery anastomosis-a nonsuture method employing tantalum rings." The Journal of Thoracic and Cardiovascular Surgery 41:378-386. Goldman, Dana, and Darius Lakdawalla. 2011. "Intellectual property, information technology, biomedical research, and marketing of patented products." In Handbook of Health Economics, 825-872. Elsevier. Hausman, Jerry A. 1978. "Specification tests in econometrics." Econometrica:1251-1271. Hausman, Jerry A, and William E Taylor. 1981. "Panel data and unobservable individual effects." Econometrica:1377-1398. Hay, Michael, David W Thomas, John L Craighead, Celia Economides, and Jesse Rosenthal. 2014. "Clinical development success rates for investigational drugs." Nature Biotechnology 32 (1):40. Kessler, David A, Janet L Rose, Robert J Temple, Renie Schapiro, and Joseph P Griffin. 1994. "Therapeutic- class wars--drug promotion in a competitive marketplace." New England Journal of Medicine 331 (20):1350-1353. Keyhani, Salomeh, Marie Diener-West, and Neil Powe. 2006. "Are development times for pharmaceuticals increasing or decreasing?" Health Affairs 25 (2):461-468. Lee, Thomas H. 2004. "“Me-too” products—friend or foe?" New England Journal of Medicine 350 (3):211-212. Leon, Martin B., Tamim M. Nazif, and Vinayak Bapat. 2018. "Interpreting National Trends in Aortic Valve Replacement: Let the Buyer Beware." Journal of the American College of Cardiology 71 (15):1628-1630. doi: 10.1016/j.jacc.2018.02.025. Lu, Z John, and William S Comanor. 1998. "Strategic pricing of new pharmaceuticals." Review of Economics and Statistics 80 (1):108-118. Makower, J, A Meer, and L Denend. 2010. FDA impact on US medical technology innovation: A survey of over 200 medical technology companies. Washington, DC, Advanced Medical Technology Association. McKinlay, John B. 1981. "From" promising report" to" standard procedure": seven stages in the career of a medical innovation." The Milbank Memorial Fund Quarterly. Health and Society 59 (3):374-411. Newhouse, Joseph P. 1992. "Medical care costs: how much welfare loss?" Journal of Economic Perspectives 6 (3):3-21. Newhouse, Joseph P. 2004. "How much should Medicare pay for drugs?" Health Affairs 23 (1):89-102. Nordhaus, William D. 1969. "An economic theory of technological change." The American Economic Review 59 (2):18-28. Sonchak, Lyudmyla. 2015. "Medicaid reimbursement, prenatal care and infant health." Journal of Health Economics 44:10-24. Stern, Ariel Dora. 2017. "Innovation under regulatory uncertainty: evidence from medical technology." Journal of Public Economics 145:181-200. Towse, Adrian , and Trevor Leighton. 1999. The changing nature of NCE pricing of second and subsequent entrants. In Risk and return in the pharmaceutical industry. London: Office of Health Economics. Ward, Michael R, and David Dranove. 1995. "The vertical chain of research and development in the pharmaceutical industry." Economic Inquiry 33 (1):70-87. Weisbrod, Burton A. 1991. "The health care quadrilemma: an essay on technological change, insurance, quality of care, and cost containment." Journal of Economic Literature 29 (2):523-552. Wertheimer, Albert, Richard Levy, and Thomas O Connor. 2001. "Too many drugs? The clinical and economic value of incremental innovations." Research in Human Capital and Development:77-118. Wilensky, Gail R. 2016. "Robotic surgery: an example of when newer is not always better but clearly more expensive." The Milbank Quarterly 94 (1):43. Wong, Chi Heem, Kien Wei Siah, and Andrew W Lo. 2019. "Estimation of clinical trial success rates and related parameters." Biostatistics 20 (2):273-286.

Wright, Jason D. 2017. "Robotic-assisted surgery: balancing evidence and implementation." JAMA 318 (16):1545-1547.

Tables and Figures Table 1: Number of CPT III Applications, Approvals, and Rejections by Year Year* (1) (2) (3) (4) (5) (6) CPT III CPT III CPT III Approvals Approvals Approvals Applications Approvals Rejections promoted to Sunsetted by Remaining as CPT I by 2019 2019 CPT III by 2019 2008 11 9 2 2 4 3 2009 19 17 2 8 4 5 2010 14 12 2 5 3 4 2011 12 10 2 3 5 2 2012 11 10 1 3 2 5 2013 15 15 0 3 5 7 2014 13 13 0 4 3 6 2015 15 15 0 1 0 14 2016 16 14 2 1 0 13 2017 18 13 5 0 0 13 2018 17 10 7 0 0 10 2019 26 24 2 0 0 24 Total 187 162 25 30 26 106

Note: The sample contains 187 CPT III applications discussed in the AMA meetings between 2008 and 2019. * Applications are classified by submission year.

Table 2. Number of CPT III Applications by Procedure Type CPT III CPT III Promoted to CPT I Sunsetted by Remaining as CPT Applications Approvals by 2019 2019 III by 2019 Panel 1: By Device Type New Device* 74 65 11 11 43 Old Device 52 50 17 11 22 No Device 18 13 2 4 7 Panel 2: By Applicant Type Industry Applicant 77 65 5 12 48 Medical Applicant 67 63 25 14 24 Notes: Sample includes 128 CPT III procedures approved between 2008 and 2017. *New devices refer to those that have not previously been used in another medical procedure and have been approved less than two years prior to CPT III application.

Table 3: Innovation Time of CPT III Procedures, in Months (Number of observations in parentheses) All By Procedure Type New Device* Old Device No Device Total: First Research to CPT III 134 98 173 164 Approval (N=128) (N=65) (N=50) (N=13) CPT III Submission to CPT III 14 14 14 14 Approval (N=128) (N=65) (N=50) (N=13) FDA Approval to CPT III 52 8 79 Approval (N=75) (N=28) (N=47) N/A 82 82 83 First Research to FDA Submission (N=75) (N=28) (N=47) N/A FDA Submission to FDA 10 14 7 Approval (N=75) (N=28) (N=47) N/A Notes: *New devices refer to those that have not previously been used in another medical procedure and have been approved less than two years prior to CPT III application

Table 4. Innovation Time of Procedures with CPT I Status, in Months (Number of observations in parentheses) All By Procedure Type New No Device* Old Device Device First Research to FDA Submission 80 (N=23) 57 (N=6) 88 (N=17) N/A FDA Submission to FDA Approval 7 (N=23) 13 (N=6) 5 (N=17) N/A FDA Approval to CPT III Application Submission 48 (N=23) 9 (N=6) 62 (N=17) N/A CPT III Application Submission to CPT III Effective 14 (N=30) 13 (N=11) 15 (N=17) 15 (N=2) 156 170 370 Subtotal (i.e., First Research to CPT III Effective) (N=30) 95 (N=11) (N=17) (N=2) CPT III Effective to CPT I Effective 38 (N=30) 34 (N=11) 41 (N=17) 30 (N=2) 194 212 400 Total: First Research to CPT I Effective (N=30) 130 (N=11) (N=17) (N=2) Notes: The sample contains 30 CPT III applications that have been approved between 2008 and 2017 and promoted by September 2019. 23 of these 30 applications involve devices that have been approved by the FDA at the time of CPT III application. *New devices refer to those that have not previously been used in another medical procedure and have been approved less than two years prior to CPT III application.

Table 5. Summary Statistics of Active CPT III Procedures between 2012 and 2017 Variable Mean SD Min Max No. of Procedure- Year Observations Sample 1: Full Sample (No. of procedures=184, No. of procedure-year observations=801) No. of Medicare Services (filling unreported 2462 13950 10 251502 801 utilization with 10) No. of Medicare Services (filling unreported 2455 13951 1 251502 801 utilization with 1) No. of Medicare Services (filling unreported 2455 13951 0 251502 801 utilization with 0) Any Medicare utilization (filling unreported 0.30 0.46 0 1 801 utilization with 0) Post-Promotion (Promoted to CPT I) 0.08 0.27 0 1 801 Tenure (No. of years since CPT III 5.67 3.36 1 15 801 approval) No Device 0.24 0.43 0 1 391 Device Approved by the year (time-variant) 0.60 0.49 0 1 391 Device Unapproved by the year (time- 0.16 0.37 0 1 391 variant) Procedure involving exclusive patented 0.53 0.50 0 1 391 medical devices (time-invariant) Procedure involving nonexclusive patented 0.23 0.42 0 1 391 medical devices (time-invariant) Sample 2: Observations for which the procedure records Medicare utilization in some years between 2012 and 2017 (No. of procedures=72, No. of procedure-year observations=319) No. of Medicare Services (filling unreported 6166 21603 10 251502 319 utilization with 10) No. of Medicare Services (filling unreported 6164 21604 1 251502 319 utilization with 1) No. of Medicare Services (filling unreported 6163 21604 0 251502 319 utilization with 0) Any Medicare utilization (filling unreported 0.75 0.43 0 1 319 utilization with 0) Post-Promotion (Promoted to CPT I) 0.18 0.38 0 1 319 Tenure (No. of years since CPT III 5.60 3.24 1 15 319 approval) No Device 0.15 0.36 0 1 167 Device Approved by the year (time-variant) 0.79 0.41 0 1 167 Device Unapproved by the year (time- 0.06 0.24 0 1 167 variant) Procedure involving exclusive patented 0.49 0.50 0 1 167 medical devices (time-invariant) Procedure involving nonexclusive patented 0.37 0.48 0 1 167 medical devices (time-invariant)

Sample 3: Observations with recorded Medicare utilization (No. of procedures=72, No. of procedure-year observations=240) No. of Medicare Services 8192 24583 11 251502 240 Post-Promotion (Promoted to CPT I) 0.24 0.43 0 1 240 Tenure (No. of years since CPT III 5.67 3.29 1 15 240 approval) No Device 0.17 0.37 0 1 134 Device Approved by the year (time-variant) 0.79 0.41 0 1 134 Device Unapproved by the year (time- 0.04 0.21 0 1 134 variant) Procedure involving exclusive patented 0.51 0.50 0 1 134 medical devices (time-invariant) Procedure involving nonexclusive patented 0.33 0.47 0 1 134 medical devices (time-invariant) Notes: Number of observations with procedure characteristics (i.e., whether involving exclusive patented medical devices, nonexclusive devices, or applied by medical societies) is less than the total because we only observe procedure characteristics for those approved after 2007.

Table 6. Main Results: Effect of CPT code promotion on Procedure Utilization Sample 1 Sample 2 Sample 3 Replacing Replacing Replacing Replacing Excluding Replacing Replacing unreported unreported unreported unreported observations with unreported unreported utilization with utilization with utilization with utilization with unreported utilization with 1 utilization with 1 10 0 10 0 utilization Dependent ( ) ( ) _ ( ) ( ) _ ( ) Variable 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(1)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(2)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷（ 3𝐷𝐷）𝑈𝑈 𝑈𝑈𝑈𝑈𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(4)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(5)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷（ 6𝐷𝐷）𝑈𝑈 𝑈𝑈𝑈𝑈𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(7)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 2.328*** 3.061*** 0.319*** 2.705*** 3.640*** 0.406*** 1.272*** (0.522) (0.743) (0.113) (0.602) (0.868) (0.136) (0.427) 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃No. of 𝐼𝐼𝑖𝑖𝑖𝑖 184 184 184 72 72 72 72 Procedures No. of Observation 801 801 801 319 319 319 240 s R-squared 0.21 0.19 0.10 0.26 0.23 0.12 0.22 Notes: ( ) represents the natural logarithm of Medicare utilization of procedure i in year t. _ represents the indicator variable for whether the procedure i records any utilization in year t. 𝑖𝑖𝑖𝑖 All regressions𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 control𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 for device approval status, procedure fixed effects, and year fixed effects. Sample𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 𝐷𝐷1 is𝑈𝑈 𝑈𝑈𝑈𝑈the𝑖𝑖𝑖𝑖 full sample. Sample 2 restricts to observations for which the procedure records Medicare utilization in some years between 2012 and 2017. Sample 3 restricts to observations with recorded Medicare utilization. Standard errors in parentheses are clustered by procedure. *** p<0.01, ** p<0.05, * p<0.1.

Table 7: CPT III procedures by Device Patent Type CPT III Exclusive Patented Nonexclusive No

approvals Device* Device Device (1) (2) (3) (4) Industry Applicants 65 60 3 2 Medical Society 58 21 26 11 Medical Physician (with Applicants 5 5 0 0 COI)** Total 128 86 29 13 Notes: Sample includes 128 CPT III procedures approved between 2008 and 2017. * A procedure is defined to involve a patented device if it a) involves a medical device; b) the medical device is made by only one firm (i.e., no competing firms) based on the referenced studies in the CPT III application; and c) the medical device company has an unexpired patent claiming the device at the time of CPT III application. ** All five physician applicants in this category disclosed Conflict of Interest (COI) and are paid consultants of medical device companies.

Figure 1: Distribution of New Medical Procedure Development Time 18 16 16 14 14

12 10 10 10 10 9 9 8 8 7 7

6 5 4 4 4 3 3 3 3 2 2 1

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 16 17 18 19 >=20

Number of years from first research to CPT III approval

Notes: Sample includes 128 CPT III procedures approved between 2008 and 2017.

Figure 2: Event study plot for the effect of CPT promotion on procedure utilization. (Full Sample) 4 Promotion from Category III to Category I CPT Codes

0 t-4 t-3 t-2 t-1 t t+1 t+2 t+3 t+4 -1 Estimated Coefficient (Mean, CI) (Mean, Coefficient Estimated 95% -2 Number of Years Relative to the CPT Promotion Year

Note: This figure presents the estimated coefficient (mean and 95% CI) of _ ( ) from Equation (2). The x- axis represents the time leads or lags from the year of CPT code promotion. The dashed line represents the time when 𝑖𝑖𝑖𝑖 𝑡𝑡 CPT code is promoted from Category III to Category I. N=801. 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸

Appendices

Appendix Figure A1: Distribution of Promotion Time 12

10 10 9

8 7 6 6

4 4 4 3 2 2 2 2 1 1 1

0 1 2 3 4 5 6 7 8 9 10 11 12 14

Number of years from CPT III approval to CPT I promotion

Note: The sample contains 52 procedures promoted to CPT I status during our time frame (i.e., between 2002 and 2019). It includes 22 procedures that are excluded from the calculation in Table 5 because they were approved as CPT III before 2008 and we do not observe their application/promotion details other than the CPT III approval date and the date of promotion to CPT I.

Appendix Figure A2: Distribution of CPT I Development Time 4

3 3 3 3 3

2 2 2 2 2

1 1 1 1 1 1 1 1 1 1 1

0 4 7 8 9 10 11 12 13 14 17 19 21 22 28 30 32 33 44

Number of years from first research to CPT I Approval

Note: The sample contains 30 CPT III applications that have been approved between 2008 and 2017 and promoted by September 2019.

Appendix Figure A3. Robustnest Test: Event study plot for the effect of CPT promotion on procedure utilization (Subsample of procedures with positive utilization) 5 Promotion from Category III to Category I CPT Codes

0 t-4 t-3 t-2 t-1 t t+1 t+2 t+3 t+4

Estimated Coefficient ()Mean, ()Mean, 95% Coefficient Estimated CI) -1

-2 Number of Years Relative to the CPT Promotion Year

Note: This figure presents the estimated coefficient (mean and 95% CI) of _ ( ) from Equation (2). The x- axis represents the time leads or lags from the year of CPT code promotion. The dashed line represents the time when 𝑖𝑖𝑖𝑖 𝑡𝑡 CPT code is promoted from Category III to Category I. The estimation uses𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 the subsample𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 of CPT III procedures with positive utilization in the Medicare utilization data. N=319.

Appendix Figure A4. Robustness Test by Random Sampling to Fill Unreported Utilization (a) Full sample—Estimated Coefficient and Standard Error .08 .08 .06 .06 .04 .04 Fraction Fraction .02 .02 0 0 .5 .525 .55 .575 .6 .625 .65 .675 .7 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3 Standard Error Estimate

(b) Subsample with some positive utilization— Estimated Coefficient and Standard Error .08 .08 .06 .06 .04 .04 Fraction Fraction .02 .02 0 0 2.3 2.4 2.5 2.6 2.7 2.8 .525 .55 .575 .6 .625 .65 .675 .7 .725 Estimate Standard Error

Note: This figure presents the distribution of the estimated coefficients and standard errors for in Equation 1. Standard errors in parentheses are clustered by procedure. 𝑖𝑖𝑖𝑖 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝐼𝐼

Appendix Table A1: Heterogeneity Analysis by Procedure Device Type Sample 1 Sample 2 Sample 3 (1) (2) (3) (4) (5) (6) 2.416*** 0.285* 2.887*** 0.560 1.350*** 0.081 (0.534) (0.164) (0.669) (0.360) (0.428) (0.308) 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃* 𝐼𝐼𝑖𝑖𝑖𝑖 1.920*** 1.949*** 1.602*** (0.468) (0.514) (0.542) 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝐼𝐼*𝑖𝑖𝑖𝑖 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝑖𝑖 3.493*** 3.413*** 1.108 (1.034) (1.042) (0.688) 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃Number𝐼𝐼𝑖𝑖𝑖𝑖 𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 of Procedures 𝑖𝑖 104 104 41 41 41 41 Number of Observations 391 391 167 167 134 134 R-squared 0.33 0.41 0.41 0.46 0.34 0.37 Notes: The dependent variable is the natural log of the number of Medicare services, replacing the unreported utilization with 10. All regressions control for device approval status, procedure fixed effects, and year fixed effects. Sample 1 is the full sample. Sample 2 restricts to observations for which the procedure records Medicare utilization in some years between 2012 and 2017. Sample 3 restricts to observations with recorded Medicare utilization. Standard errors in parentheses are clustered by procedure. *** p<0.01, ** p<0.05, * p<0.1.

Appendix Table A2. Robustness Test—Adding Interaction between Tenure and Log Time Trend Sample 1 Sample 2 Sample 3 Replacing Replacing Replacing Replacing Excluding Replacing Replacing unreported unreported unreported unreported observations with unreported unreported utilization with utilization with utilization with utilization with unreported utilization with 1 utilization with 1 10 0 10 0 utilization Dependent ( ) ( ) _ ( ) ( ) _ ( ) Variable 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(1)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(2)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷（ 3𝐷𝐷）𝑈𝑈 𝑈𝑈𝑈𝑈𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(4)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(5)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷（ 6𝐷𝐷）𝑈𝑈 𝑈𝑈𝑈𝑈𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(7)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 2.343*** 3.040*** 0.302*** 2.630*** 3.437*** 0.351** 1.271*** (0.527) (0.753) (0.115) (0.610) (0.899) (0.145) (0.433) 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃No. of 𝐼𝐼𝑖𝑖𝑖𝑖 184 184 184 72 72 72 72 Procedures No. of Observation 801 801 801 319 319 319 240 s R-squared 0.26 0.24 0.14 0.35 0.33 0.22 0.38 Notes: ( ) represents the natural logarithm of Medicare utilization of procedure i in year t. _ represents the indicator variable for whether the procedure i records any utilization in year t. 𝑖𝑖𝑖𝑖 All regressions𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 control𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 for device approval status, procedure fixed effects, and year fixed effects. Sample𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 𝐷𝐷1 is𝑈𝑈 𝑈𝑈𝑈𝑈the𝑖𝑖𝑖𝑖 full sample. Sample 2 restricts to observations for which the procedure records Medicare utilization in some years between 2012 and 2017. Sample 3 restricts to observations with recorded Medicare utilization. Standard errors in parentheses are clustered by procedure. *** p<0.01, ** p<0.05, * p<0.1.

Appendix Table A3. Robustness Test—Excluding Procedures with Unapproved Devices at the Time of CPT III Approval Sample 1 Sample 2 Sample 3 Replacing Replacing Replacing Replacing Excluding Replacing Replacing unreported unreported unreported unreported observations with unreported unreported utilization with utilization with utilization with utilization with unreported utilization with 1 utilization with 1 10 0 10 0 utilization Dependent ( ) ( ) _ ( ) ( ) _ ( ) Variable 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(1)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(2)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷（ 3𝐷𝐷）𝑈𝑈 𝑈𝑈𝑈𝑈𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(4)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(5)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷（ 6𝐷𝐷）𝑈𝑈 𝑈𝑈𝑈𝑈𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(7)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 2.266*** 3.056*** 0.343** 2.856*** 3.887*** 0.447*** 1.040** (0.660) (0.952) (0.138) (0.742) (1.084) (0.165) (0.495) 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃No. of 𝐼𝐼𝑖𝑖𝑖𝑖 148 148 148 61 61 61 61 Procedures No. of Observation 687 687 687 281 281 281 209 s R-squared 0.16 0.15 0.10 0.21 0.21 0.13 0.10 Notes: ( ) represents the natural logarithm of Medicare utilization of procedure i in year t. _ represents the indicator variable for whether the procedure i records any utilization in year t. 𝑖𝑖𝑖𝑖 All regressions𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 control𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 for device approval status, procedure fixed effects, and year fixed effects. 𝐷𝐷Sample𝐷𝐷𝐷𝐷𝐷𝐷 𝐷𝐷1 is𝑈𝑈 𝑈𝑈𝑈𝑈the𝑖𝑖𝑖𝑖 full sample. Sample 2 restricts to observations for which the procedure records Medicare utilization in some years between 2012 and 2017. Sample 3 restricts to observations with recorded Medicare utilization. Standard errors in parentheses are clustered by procedure. *** p<0.01, ** p<0.05, * p<0.1.

Appendix Table A4. Robustness Test—Effect of CPT code promotion on Procedure Utilization Using the Year before CPT I publication year as the promotion year. Sample 1 Sample 2 Sample 3 Replacing Replacing Replacing Replacing Excluding Replacing Replacing unreported unreported unreported unreported observations with unreported unreported utilization with utilization with utilization with utilization with unreported utilization with 1 utilization with 1 10 0 10 0 utilization Dependent ( ) ( ) _ ( ) ( ) _ ( ) Variable 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(1)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(2)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷（ 3𝐷𝐷）𝑈𝑈 𝑈𝑈𝑈𝑈𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(4)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(5)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷（ 6𝐷𝐷）𝑈𝑈 𝑈𝑈𝑈𝑈𝑖𝑖𝑖𝑖 𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈(7)𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 𝑖𝑖𝑖𝑖 1.555*** 2.225*** 0.291** 1.917*** 2.849*** 0.405** 0.655* (0.470) (0.687) (0.133) (0.573) (0.827) (0.162) (0.343) 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃No. of 𝐼𝐼𝑖𝑖𝑖𝑖 184 184 184 72 72 72 72 Procedures No. of Observation 801 801 801 319 319 319 240 s R-squared 0.08 0.05 0.11 0.13 0.12 0.09 0.16 Notes: ( ) represents the natural logarithm of Medicare utilization of procedure i in year t. _ represents the indicator variable for whether the procedure i records any utilization in year t. 𝑖𝑖𝑖𝑖 All regressions𝑙𝑙𝑙𝑙 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 control𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈 for device approval status, procedure fixed effects, and year fixed effects. 𝐷𝐷Sample𝐷𝐷𝐷𝐷𝐷𝐷 𝐷𝐷1 is𝑈𝑈 𝑈𝑈𝑈𝑈the𝑖𝑖𝑖𝑖 full sample. Sample 2 restricts to observations for which the procedure records Medicare utilization in some years between 2012 and 2017. Sample 3 restricts to observations with recorded Medicare utilization. Standard errors in parentheses are clustered by procedure. *** p<0.01, ** p<0.05, * p<0.1.

Appendix Table A5: Medicare Utilization (Measured by Number of Services) of Top 5 Utilized New Medical Procedures# AMA Approval Year (Category Procedure Name III CPT Code)a 2013 2014 2015 2016 2017 External ECG Monitoring 2012 28,161 56,291 108,442 176,264 251,502 Surface Electronic High Dose Rate Brachytherapy 2015 - - - 26,452 36,247 Visual Field Assessment with Real Time Data Analysis 2014 - - 415 14,651 33,752 Subcutaneous Implantable Defibrillator 2013 122 1,712 4,109* 5,810* 6,328* Sacroiliac Joint Stabilization 2013 - 175 1,910* 2,810* 3,704* Left Atrial Appendage Closure 2011 151 71 288 3,223 8,801* TAVR 2011 15,823* 26,472* 37,882* 54,449* 67,783* Circulating Tumor Cells (CTC) Enumeration 2011 6,541* 4,031* 2,115* 1,743* 546* Ultrasound Guided Facet Injection 2010 6048 3,398 N/A N/A N/A Unattended Sleep Study 2009 9,003* 11,490* 13,115* 16,145* 19,174* Intrafraction Target Tracking 2009 24,377 23,466 N/A N/A N/A

Notes: Cells show Medicare utilization of each procedure in each year, measured by the number of unique Medicare beneficiary/provider interactions. Utilization data are from Medicare Provider Utilization and Payment Data: Physician and Other Supplier. # New medical procedures are defined as those approved as Category III CPT codes within 5 years of the corresponding year. Procedures are selected so that the top 5 new procedures by total utilization are included for each year. Numbers in Bold are the top5-utilized procedures in the corresponding year. Shaded cells correspond to procedures that were introduced more than 5 years before the corresponding year (so are not considered ‘new’). * The procedure has been promoted to Category I CPT status in the corresponding year. a CPT codes are normally available to use in the following year of the approval year.

Appendix Table A6: Medicare Utilization (Measured by Number of Medicare Beneficiaries) of Top 5 Utilized New Drugs# FDA Approval Drug Name Year 2013 2014 2015 2016 2017 ELIQUIS 2012 46,920 204,210 481,422 826,969 1,142,004 BREO ELLIPTA 2013 632 44,744 133,609 293,833 533,708 MYRBETRIQ 2012 66,432 147,553 213,641 296,934 394,967 LINZESS 2012 53,657 144,002 204,280 268,598 321,437 INVOKANA 2013 18,624 91,499 194,566 233,132 231,835 XARELTO 2011 416,543 650,370 727,624 807,820 951,753 TRADJENTA 2011 105,342 155,144 227,831 276,586 301,919 PRADAXA 2010 250,767 238,057 221,745 229,987 231,294 DEXILANT 2009 310,989 305,373 312,967 313,985 299,688 COLCRYS 2009 431,070 465,482 238,512 144,750 185,601 BYSTOLIC 2009 401,397 399,956 366,945 346,811 338,263 Notes: Cells show Medicare utilization of each drug in each year, measured by the number of Medicare beneficiaries. Utilization data are from Medicare Provider Utilization and Payment Data: Part D Prescriber. # New drugs are defined as those approved within 5 years of the corresponding year. Drug are selected so that the top 5 new drugs by total utilization are included for each year. Numbers in Bold are the top5-utilized drugs in the corresponding year. Shaded cells correspond to drugs that were introduced more than 5 years before the corresponding year (so are not considered ‘new’).

Appendix Table A7: Payments for New Procedures versus New Drugs Payments ($ million) Top 5-Utilized New Procedures* Top 5-Utilized New Drugs** 2013 679 1,848 2014 1,150 2,432 2015 1,668 3,550 2016 120 3,838 2017 151 5,977 5-yr Total 3,768 17,645 *Payments for new procedures are from Medicare Provider Utilization and Payment Data: Physician and Other Supplier. ** Payments for new drugs are from Medicare Utilization and Payment Data: Part D Prescriber.

Appendix Table A8: Year of Introduction of Top 15 Utilized Medical Procedures by Medicare Beneficiaries in 2017 Medicare Beneficiary/Provider Year of Procedure Interactions (Million) Medicare Spending ($ Million) Introduction* X-ray 53 629 1896 Collection of Venous Blood by Venipuncture 24 70 N/A CT Scan 22 1,320 1972 Removal of Skin Lesions (Benign and Malignant) 11.73 571 1938 Ultrasound 11 484 1956 Cataract Removal and Lens Insertion 9.5 6,315 1967 Biopsy 9.0 1,057 1875 Endoscopic Diagnostic Examination 3.0 399 1853 Removal of Ear Wax 1.4 41 N/A Complex Wound Repair 0.72 157 N/A Knee Repair (incl. Replacement) 0.56 6,750 Early 1970s Drainage of Abscess/Pilonidal Cyst 0.49 47 N/A Dialysis (Outpatient)** 0.40 (approx.) 11,400 (approx.) 1943 Prosthetic Hip Replacement 0.28 4,089 1940 Endotracheal Intubation 0.28 32 1878 Notes: Utilization and spending data are from Medicare Provider Utilization and Payment Data: Physician and Other Supplier, and Medicare Provider Utilization and Payment Data: Outpatient. * See Appendix B for sources for year of introduction. ** Dialysis estimates are based on Medicare Payment Advisory Commission (2019).

Appendix Table A9: Year of Introduction of Top 15 Utilized Drugs by Medicare Beneficiaries in 2017 Medicare Part D Year of Initial Medicare Spending ($ Drug (Active Ingredient) Typical Use Beneficiaries FDA Approval* Million) (Million) (Brand) Hypercholesterole 1996 ATORVASTATIN CALCIUM 10.7 878 mia (LIPITOR) 2002 LEVOTHYROXINE SODIUM Hypothyroidism 8.4 1,120 (LEVO-T) Hypertension, 1987 AMLODIPINE BESYLATE 8.3 288 Angina (NORVASC) Hypertension, 1988 LISINOPRIL Heart Failure, 8.2 263 (ZESTRIL) Kidney Disease HYDROCODONE/ACETAMI 1997 Pain 6.9 508 NOPHEN (NORCO) Gastroesophageal 1989 OMEPRAZOLE 6.9 395 Reflux Disease (PRILOSEC) 1995 METFORMIN HCL Diabetes 6.8 677 (GLUCOPHAGE ) 1995 AZITHROMYCIN Bacterial Infections 6.1 73 (ZITHROMAX) 1991 SIMVASTATIN Heart Disease 5.9 223 (ZOCOR) Arthritis, Blood Disorders, 1955 PREDNISONE 5.8 119 Breathing (RAYOS) problems, etc. 1993 GABAPENTIN Seizures 5.8 549 (NEURONTIN) 1981 ALBUTEROL SULFATE Asthma 5.6 915 (PROAIR) Edema (Caused by 1968 FUROSEMIDE Heart, Kidney and 5.5 141 (LASIX) Liver Disease) 1974 AMOXICILLIN Bacterial Infections 5.1 32 (AMOXIL) Hypertension, 1995 LOSARTAN POTASSIUM Heart Failure, 4.9 226 (COZAAR) Kidney Disease Notes: Utilization and spending data are from Medicare Provider Utilization and Payment data: Part D Prescriber, and year of FDA approval is from the Drugs@FDA database. * Year of FDA Approval gives the earliest date of FDA approval for a product containing the active ingredient.

Appendix B: Year of Introduction of Top-utilized Medical Procedures shown in Appendix Table A6. X-ray: https://www.nde- ed.org/EducationResources/CommunityCollege/Radiography/Introduction/history.htm#:~:text=In%20Jun e%201896%2C%20only%206,pictures%20of%20metals%20were%20produced CT Scan: https://www.imaginis.com/ct-scan/brief-history-of- ct#:~:text=Tomography%20is%20from%20the%20Greek,Cormack%20of%20Tufts%20University%2C%20 Massachusetts Removal of Malignant Lesions: https://en.wikipedia.org/wiki/Mohs_surgery Ultrasound: https://www.livescience.com/32071-history-of-fetal- ultrasound.html#:~:text=When%20it%20was%20invented%3F,detect%20industrial%20flaws%20in%20ship s Cataract Removal and Lens Insertion: https://eyewiki.aao.org/History_of_Cataract_Surgery Biopsy: https://pubmed.ncbi.nlm.nih.gov/7975522/#:~:text=The%20term%20%22biopsy%22%20was%20introduc ed,in%201875%20by%20M.%20M.%20Rudnev Endoscopic Diagnostic Examination: https://www.olympus- global.com/technology/museum/endo/?page=technology_museum Knee Repair (incl. Replacement): https://www.intechopen.com/books/arthroplasty-update/the- evolution-of-modern-total-knee-prostheses Dialysis (Outpatient): https://www.dpcedcenter.org/news-events/news/a-brief-history-of- dialysis/#:~:text=The%20history%20of%20dialysis%20dates,patient%20suffer%20from%20kidney%20failu re Prosthetic Hip Replacement: https://en.wikipedia.org/wiki/Hip_replacement#:~:text=On%20September%2028%2C%201940%20at,the %20cobalt%2Dchrome%20alloy%20Vitallium Endotracheal Intubation: https://pubmed.ncbi.nlm.nih.gov/16400793/#:~:text=In%20the%20early%201870's%2C%20Trendelenbur g,elective%20endotracheal%20intubation%20for%20anesthesia.&text=In%201913%20the%20first%20anest hetic,the%20Magill%2C%20Miller%20and%20Macintosh